Ontology Alignment Revisited

Semantic Web 1 (0) 1–5 1 IOS Press

1 1 2 2 3 3 4 Ontology Alignment Revisited: A 4 5 5 6 Bibliometric Narrative 6 7 7 8 Majid Mohammadi a,*, Amir Ebrahimi Fard a 8 9 a Delft University of Technology, The Netherlands 9 10 10 11 11 12 12 13 13 14 Abstract. Ontology alignment is an important problem in the Semantic Web with diverse applications in various disciplines. 14 15 This paper delineates this vital field of study by analyzing a core set of research outputs from the domain. In this regard, the 15 16 related publication records are extracted for the period of 2001 to 2018 by using a proper inquiry on the well-known database 16 Scopus. The article details the evolution and progress of ontology alignment since its genesis by conducting two classes of 17 17 analyses, namely, semantic and structural, on the retrieved publication records from Scopus. Semantic analysis entails the overall 18 18 discovery of concepts, notions, and research lines flowing underneath ontology alignment, while the structural analysis provides 19 a meta-level overview of the field by probing into the collaboration network and citation analysis in author and country levels. 19 20 In addition to these analyses, the paper discusses the limitations and puts forward lines for the further progress of ontology 20 21 alignment. 21 22 22 23 Keywords: ontology alignment, bibliometrics, scientometric 23 24 24 25 25 26 26 27 1. Introduction In this regard, a pre-processing strategy is required to 27 28 align the ontologies of these heterogeneous informa- 28 29 The Semantic Web is an extension of the World tion systems, after which they can interact with each 29 30 Wide Web which aims to provide metadata for ma- other. This strategy is called ontology alignment (also 30 31 chines so that they can further understand and construe called ontology mapping and ontology matching). 31 32 the published information and data on the Web. The The necessity of having a tool to automatically align 32 33 consequence of having such metadata is that comput- two different ontologies was recognized in the late 33 34 ers can also make reasonable interpretations, ideally 90s and early 2000 [73–75, 98]. Since the heterogene- 34 35 identical to how humans process and understand the ity problem was quite epidemic and was a stumbling 35 36 information. The main key to providing such informa- block in the way of interoperability, it soon found its 36 37 tion for computers is ontologies, with which one can way in many applications such as agent communica- 37 38 model the underlying objects of a domain along with tion in agent-based modeling [98], ontology merging 38 39 their interrelations. The design of ontologies is thor- [74, 75], data integration [80], business-to-business e- 39 40 oughly subjective and is primarily reliant on the vi- commerce [76], ontology development and visioning 40 41 sion of the creator. Since humans potentially consider [16], database evolution [17, 58], web service discov- 41 42 different aspects of a domain, or they might use dif- ery [23], to name just a few. Due to the diverse ap- 42 43 ferent terminology for similar concepts, the ontologies plications of ontology alignment, many research stud- 43 44 are not uniquely defined and are distinct from each ies have been dedicated to resolving the heterogene- 44 45 other, even those from one particular domain. If infor- ity among information systems, and new problems are 45 46 mation systems using these ontologies are assumed to modeled to be solved by alignment techniques. 46 47 work independently, the difference in ontologies does In recent two decades, tremendous efforts have been 47 48 not cause any issue. However, the issue emerges when taken to further improve the field of ontology align- 48 49 these systems want to interoperate and exchange data. ment [26, 45, 67–70, 77]. As a result of such efforts, 49 50 there are several valuable materials available for ontol- 50 51 *Corresponding author. E-mail: [email protected]. ogy alignment researchers and others who want to uti- 51

1 lize or get acquainted with alignment techniques. For a parameter-free measure that quantifies the extent to 1 2 instance, there is a book containing the fundamentals which a specific paper can be considered an SB. 2 3 of ontology alignment and recent advances made in The other line of studies in bibliometrics calibrates 3 4 this field [26]. Further, there are some useful reviews research activities to provide insights regarding dy- 4 5 and surveys [77, 94] and some other stating the cur- namic and vital influential factors behind scientific re- 5 6 rent state and future challenges of ontology alignment search. Citation analysis [38, 88], co-authorship analy- 6 7 [89]. Although these materials are essential and help sis [35, 85, 88], and co-occurrence word analysis [15] 7 8 researchers get familiar with the notions of ontology are prevalent in this application domain of bibliomet- 8 9 alignment, they do not provide an overview of the field. rics. For instance, in the study carried out by Bromham 9 10 Such an overview is not only critical to those who want et al. [8] on The Australian Research Council’s grant 10 11 to get familiar with the domain, but it is also vital for proposal data, they studied the relationship between 11 12 researchers in this field to track its progress and find research interdisciplinarity and the chance of winning 12 13 new ways to further improve it. In addition, such stud- grants and discovered the higher the degree of interdis- 13 14 ies allow science policymakers to get a comprehensive ciplinarity, the lower the probability of being funded. 14 15 picture of the field. This enables them to assess the Also, in another research focusing on collaboration 15 16 scientific advancement of this subject domain. Assess- in the field of Genomics, Petersen et al. [79] discov- 16 17 ment is a vital element in decision-making as it gives a ered cross-disciplinary research draw more attention 17 18 realistic picture of the current situation and prevent us and get more citations correspondingly. One type of 18 19 from over- or under-estimation. Therefore, to be sure research that falls into the same line of research is the 19 20 the right decisions are made for this field to proceed in bibliometric analysis of scientific fields. For instance, 20 21 21 the right path, assessment is essential. In this regard, Frank et al. [32] studied the bibliometric evolution of 22 AI research and its related fields since 1950. 22 this article utilizes the bibliometric analysis to provide 23 On the other side of the spectrum, bibliometrics is 23 a bird’s-eye view to the interdisciplinary domain of on- 24 utilized to address much broader goals. Some studies 24 tology alignment. 25 use bibliometrics data or analysis for answering ques- 25 Bibliometrics is a quantitative approach to study sci- 26 tions which are not for the purpose of scientific ac- 26 entific activities. At its most fundamental level, biblio- 27 tivities evaluation. This is a very recent approach to- 27 metrics aims to unveil the latent dynamics of scientific 28 ward bibliometrics, which can provide an opportunity 28 research and analyze its key influential factors. Con- 29 for other disciplines to benefit the tools and techniques 29 tent, citation, and collaboration analyses are among 30 developed in this field. For instance, Candia et al. [10] 30 the commonly-practiced techniques using bibliomet- 31 studied the problem of collective memory decay us- 31 rics analysis. In this regard, many research fields use 32 ing multiple datasets including American Physical So- 32 these techniques to delineate the importance of their 33 ciety (APS) papers and the United States Patent and 33 34 fields, the impact of the lead researchers, or gauge the Trademark Office (USPTO) patents. In the other work, 34 35 impact of a particular research output [61]. In recent Guimera et al. [36] studied the self-assembly of cre- 35 36 years, the bibliometric analysis has drawn a lot of at- ative teams in the collaboration network using empiri- 36 37 tention and is applied to major bibliometrics data en- cal study over a bibliometric dataset constitutes of 50 37 38 gines. Bibliometrics covers a broad spectrum of ap- years records of recognized journals in social psychol- 38 39 plication domains [31]. Some of the studies in biblio- ogy, ecology, economics, and astronomy. In another 39 40 metrics have more methodological orientations and try research, Liu et al. [60] studied the phenomenon of a 40 41 to scrutinize the existing bibliometric measures, e.g., hot streak for individuals career. They conducted this 41 42 citation and impact factor, or to come up with new study by combining over 20,000 researcher profiles in 42 43 ones. For instance, Chorus et al. [12] defined a met- Google Scholar and Web of Science. Ebrahimi Fard 43 44 ric for self-citation and studied trends in impact factor et al. [21] also used bibliometric analysis to study the 44 45 biased self-citations of scholarly journals. In the other readiness of academia amid a war with the diffusion of 45 46 research, Thelwall et al. [93] made a comparison for fake-news in social media. 46 47 11 altmetrics in Web of Science to understand the re- This article brings forth a bibliometric approach 47 48 lationship between real citations and alternative met- to analyze the growth and advancement of ontology 48 49 rics in social media. In another prominent study, Ke alignment. In this regard, we searched Scopus to ex- 49 50 et al. [52] made a large-scale analysis of the sleeping tract the research outputs regarding ontology align- 50 51 beauty (SB) phenomenon in science and introduced ment. We based the bibliometric analysis on the Sco- 51 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative 3

1 pus data since other databases such as Web of Science 2. Research Methodology 1 2 (WoS) do not index the ontology matching workshop, 2 3 the primary venue in this field. We retrieve and ana- In this section, we first discuss the research strat- 3 4 lyze around 2,975 research outputs from Scopus, in- egy that is employed to extract ontology alignment 4 5 cluding articles, conference papers, book chapters, and research outputs from two well-known databases, 5 6 reviews. namely Web of Science (WoS) and Scopus. We then 6 7 We carry out two classes of bibliometric analyses explain the tools and methods that are used for the 7 8 on the retrieved articles from Scopus. The first one analysis of extracted bibliometric data in further sec- 8 9 is semantic analysis concerning the overall discovery tions. 9 10 of concepts, notions, and research lines flowing un- 10 11 2.1. Ontology Alignment Bibliometric Search 11 derneath the scientific disciplines. In this analysis, we 12 Approach 12 use latent Dirichlet allocation (LDA) [5] to model the 13 13 topics underlying the ontology alignment bibliomet- 14 Bibliometric approaches aim at the quantitative 14 ric data. To do so, the title, abstract, and keywords of 15 analysis of the research outputs, such as publications 15 each document were subjected to LDA, and six top- 16 and patents, in order to comprehend and track the 16 ics were extracted accordingly. Although the topics 17 scale, direction, and the innovation of a field. The ma- 17 18 are extracted based merely on the words and their fre- jor prerequisite for such an analysis is to find the rel- 18 19 quency in each document, the extracted topics are in- ative research items according to which the analysis 19 20 terestingly meaningful and delineate different appli- could be performed. In this regard, there are several 20 21 cations to which ontology alignment can be applied well-known databases such as Scopus and Thomson 21 22 or the problems it can address. Another analysis in Reuters Web of Science, from which the research items 22 23 this category is thematic, in which we show the shares can be retrieved by proper queries. 23 24 of ontology alignment to top-cited articles and top For the bibliometric analysis, there are several stan- 24 25 percentile journals. Also, the fundamental disciplines dard ways to extract the pertinent research items to 25 26 contributing to ontology alignment are discussed in a problem/domain. Index-based methods [13] use the 26 27 this analysis. In addition to semantic analysis, we per- categories already defined by the publication database 27 28 form structural analysis in order to obtain a meta-level and retrieve the research outputs accordingly. The ap- 28 29 overview of the field. We break the structural analy- proach is simple, but the search is restricted to the 29 30 sis into two categories. First, we analyze the collabo- indices created by the journals. In particular, we ob- 30 31 rations between different authors and countries in on- served that there is no particular index for ontology 31 32 tology alignment based on their co-authorships in the matching in several prestigious publishers such as 32 33 bibliometric data. Second, we gauge the impact of re- IEEE and ACM. Another approach is based on citation 33 34 searchers and countries by analyzing their number of and co-citation [100], wherein one first needs to find 34 35 published articles and their number of citations and a core corpus of research outputs that everyone agrees 35 36 36 further visualizing their citation networks. The analy- upon. The basic corpus of publications then evolves by 37 using its citations and co-citations. The major draw- 37 sis of bibliometric data helped us address some current 38 back of this technique is that it is difficult to replica- 38 issues in the field and also provide some solutions for 39 tion, and there is no consensus on the interpretation of 39 its further improvement. 40 citations and co-citations. For ontology alignment, in 40 The remainder of this article is structured as follows. 41 particular, finding a core amount of publications which 41 Section 2 is dedicated to the methodology by which 42 everyone agrees upon is not easy to acquire. One po- 42 43 we retrieve ontology alignment research outputs and tential way would be to use the papers published in the 43 44 tools that are used to analyze them. Semantic analysis ontology matching workshop, but the number of arti- 44 45 is covered in Sections 3 and 4, where the topic analysis cles in the workshop is quite restricted so that the fi- 45 46 is presented in the former and the thematic analysis nal corpus would not include the exhaustive set of all 46 47 is discussed in the latter. Section 5 is devoted to the publications for this problem. Another way to get the 47 48 collaboration analysis, and the impact analysis at the bibliometric data is to detect a set of journals dedicated 48 49 author and country levels are explained in Section 6. to a domain and analyze their published articles [59]d. 49 50 We conclude the paper and discuss the lesson learned For ontology alignment, unfortunately, there is no par- 50 51 from the analyses in Section 7. ticular journal to conduct the analysis. On top of that, 51 4 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative

1 Table 1 1 Total Relevant Irrelevant 2 Four steps for filtering the ontology alignment research outputs. is the number of items at the beginning of each step, and 2 denote the number of items that are flagged as related and unrelated to ontology alignment, respectively, and Uncertain is the number of items 3 3 that could not be flagged as either relevant or irrelevant so that they are passed to the next phase. The search query for retrieving data from Scopus 4 is: TITLE-ABS-KEY ( "ontology matching" OR "ontology Alignment" OR "ontology mapping" OR "OAEI" ). 4 5 5 Total Relevant Irrelevant Uncertain Description 6 6 1 - - - 3289 Retrieving bibliometric data from Scopus 7 7 2 3289 1820 225 1244 Inspecting the publication items by revising the title only 8 8 3 1244 1094 53 97 Inspecting the publication items by revising the abstract only 9 9 4 97 61 36 0 Inspecting the whole paper 10 10 sum - 2975 314 - 11 11 12 12 ontology alignment is interdisciplinary by nature since ducted an inquiry in WoS by searching the identified 13 13 it is used as a pre-processing strategy in many circum- keywords in the title, abstract, and keywords of the re- 14 14 stances and has thus diverse applications. As a result, search outputs. The result of the search included 1,536 15 15 the research outputs are not restricted to a specific jour- articles spanning from 1999 until November 2018. The 16 16 nal or domain. 1,536 articles were processed in three different phases 17 17 One of the most popular yet straightforward meth- to leave out the articles that are not pertinent to the 18 18 19 ods is to use several expert-defined keywords based domain. In the beginning, the title of papers was con- 19 20 on which the research outputs are retrieved [56, 81]. sidered since most of the related works to ontology 20 21 This method is semi-automatic since the results of the alignment could be easily detected. In this phase, 1,166 21 22 search are then reviewed by an expert to exclude the items were identified as relevant or irrelevant, and 370 22 23 irrelevant items from the analysis. After inspecting the items were passed to the second phase. In the second 23 24 keywords of top 50 cited research outputs in this do- phase, the abstract of the papers was regarded in which 24 25 main and further discussion with the experts in this 316 articles were recognized as relevant, and the re- 25 26 domain, we arrived at three main keywords: “ontol- mainder 54 were passed to the third phase. In the final 26 27 ogy alignment", “ontology matching", and “ontology stage, the 54 articles were thoroughly inspected and 27 28 mapping". The keyword "ontology" alone refers to a the papers were classified as relevant and irrelevant. 28 29 more general concept in the Semantic Web and add re- In total, 1,420 research items were labeled as relevant 29 30 search items that are irrelevant to ontology alignment. to ontology alignment and the rest article were elimi- 30 31 Thus, we need to conduct the search based on the key- nated. 31 32 word “ontology alignment", which is interchangeably After rigorous examinations of the remaining arti- 32 33 referred to as “ontology matching" or “ontology map- cles, we realized that the research items regarding the 33 34 ping" as well. Thus, these terms should be considered ontology matching workshop are not indexed by WoS. 34 35 for searching the databases. We further realize that Since this workshop is the essential venue of this do- 35 36 the ontology alignment evaluation initiative (OAEI) is main, we refused to continue the analysis based on 36 37 also essential since it might also add some research WoS data. Therefore, we conducted the same search 37 38 items. Since the research items that contain “ontology strategy in Scopus and realized that the items recov- 38 39 alignment evaluation initiative" are completely covered by this database include the articles from the on- 39 40 ered by articles retrieved solely by the keyword “ontology matching workshop as well. The inquiry in Sco- 40 41 tology alignment", this term is redundant. However, pus retrieved 3,289 articles from 2001 up until 2018. 41 42 “OAEI" must be used as another keyword. Since the Although it does not index several papers from the late 42 43 use of keywords to get the articles could retrieve a con- 90s and early 2000 [73, 75], it includes all the paper 43 44 siderable number of articles in ontology alignment, we from the ontology matching workshop. Since the num- 44 45 use this technique to get ontology alignment research ber of articles that are not indexed by Scopus is not sig- 45 46 outputs. nificant, especially compared to WoS, we use Scopus 46 47 We used four keywords to retrieve the research out- data for further analysis. The retrieved articles from 47 48 puts related to ontology alignment and conducted the Scopus underwent the same procedure as Web of Sci- 48 49 inquiry on Web of Science (WoS). Further, the identi- ence articles in order to discard the irrelevant papers. 49 50 fied research items need to be processed by an expert to After conducting the three phases of processing the re- 50 51 verify if they are relevant to the domain. First, we con- search items, 2,975 articles are labeled as relevant to 51 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative 5

1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 Fig. 1. The workflow of the analyses conducted in this paper. 29 30 30 31 ontology alignment. Table 1 tabulates the taken steps disciplines contribute to the interdisciplinary ontology 31 32 for obtaining and cleaning the bibliometric data from alignment. 32 33 Scopus. Further analyses are performed on the remain- The second type of analysis is structural analysis, 33 34 34 ing items after taking these steps. that is divided into two subcategories. We initially an- 35 35 36 alyze the collaborations between authors and countries 36 2.2. Tools and Methods for Bibliometric Analysis 37 worldwide and scrutinize the level of international and 37 38 academic-corporate collaborations over the last few 38 39 In this section, we explain the methods and tools years. We then probe into the contributions and im- 39 40 that are used to analyze the 2,975 ontology alignment pacts of authors and countries in ontology alignment. 40 41 research outputs extracted from Scopus. We conduct For these analyses, we use VOSviewer [95], SciVal1, 41 two levels of analysis regarding the structural and se- 42 and also Gephi for network visualization [3]. Some of 42 mantic of the extract articles. For the semantic analy- 43 the analyses of this section are limited to recent six 43 sis, we use latent Dirichlet allocation (LDA) [5], which 44 years due to the fact that more bibliometric metadata 44 45 is a method for topic modeling. This analysis aims to 45 are only available in the recent years. Figure 1 displays 46 find the underlying topics in ontology alignment based 46 the workflow of this article, from data acquisition to 47 on the articles published in this domain in the period 47 48 2001-2018. We use the LDA implementation in MAT- their processing using different tools. 48 49 LAB to analyze the articles. We also conduct a the- 49 50 matic analysis where we discuss the number of pub- 50 51 lications overall and in top journals along with the 1https://www.scival.com 51 6 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative

Topic 1 Topic 2

1 demonstrate supply 1 interoperation function manufacture objective compose mean generate page artificial collect development reconciliation heterogeneous automatic build automated 2 depend collaborative architecture terminology software currently 2 strength minimal 2010 infrastructure annotation architecture selection emerge ontologybased vocabulary enable enables environment repair standard framework research discover document engine key protocol specific 3 computer code lead ontologybased benefit descriptiondynamic maintenance 3 event environment directory customer gap collaboration discussed maintain framework automatically support collaborative recommendation record product among major owl appropriate enterprise ecommerce chain order patient reserve manage retrieval evolution user distribute group target topic cloud exist translation 4 heterogeneous digital original component available workflow 4 educational library context message goal example search discovery software project become order thus 2009 level modelling process grid study provide fault reuse new provide implement 5 dynamic module case remove analysis mechanism resourceweb flexible 5 shareagentieee layer various rough local achieve argument content intelligent implement negotiation acquisition intelligence ieee communication support make integrate electronic global migration relevant 6 heterogeneity interactionpower implementation compositionservicedevelop management 6 lattice exchange methodology allow health formal diagnosis platform describe adaptation subject multiagent medical clinical provider request repository translate metamatching ontological efficient input copyright runtime focus methodology intelligent distribute plan evolve 7 annotate partner development 7 preserve mechanism argumentation manufacturing 8 8 Topic 3 Topic 4 expansion requires 9 construction discussed flow construction 9 generation simulation security relationship description organization describes scheme healthcare domainspecific understanding vocabulary functional way research store statistical natural metadata achieve transform transformation 2009 10 access purpose resolve share 10 creation representation define participate test relies social participation widely form implementation campaign capture theory requirement control foundation distribute queries cultural element virtual ieee build modeling retrieve querying realize specify logmap describe 11 operation document relational agentbased 11 provide accord semiautomatic platform include give provide first detail profile managementresponse merging illustrate connect mediation big mean roleenable accurate create schema step specific specification rulestructure required three develop establish contest table 12 inference represent disease 12 semantics interface source evaluate engineering evidence benchmark build initiative processeven facilitate ontological common term query rdf 2007 report none category integrate express 2012 owl year tag implement 13 device file access publish 2011 13 among text business describedevelopment xml inference smart internet general track object formal 2010 sparql level analysis ontologybased conceptual mechanism conference aim apply heterogeneous exploitation activity reuse demand anatomy last rule 14 solve analysis equivalent database 14 include well discusses project article engineer develop type definition dataset format prototype heterogeneity difference unified complex heterogeneous collection practice bridge 15 expression particularly obtain association 15 probability improvementclassifier

16 Topic 5 Topic 6 16

number anchor consistency researchers name 17 crosslingual automatically sense increase 17 furthermore amount partition general retrieval selection start thesaurus component several linking experimental effective wordnet solutionstandard correspondence structural size external 18 current effectiveness 18 semantics thus propertyautomatically science machine label global score multistrategy exist repository background well future reserve corpus many automated multilingual reasoning recent 2012 generate construct content link framework logical study 19 open express 19 interest key biomedical various logic contain associate conclusion target tree learning multiple detect often art survey instancebased review web due term real authors identify manydevelop novel linguistic classification extract object concepts 20 integrate issue focus crucial world contain finally lexical 20 block input area lod allow dataset new word life class owl researchrealworld task still state performance learnsensor cluster probabilistic pattern exist large scale field class individual type case exploit year configuration mining 21 far structure small 21 number linked largerelated check bayesian description datasets extend graphgene property important detection annotation well analysis play reason address relationship pair four potential study challenge hierarchical task analyze feature provide identify apply 22 useful basis 22 particular automatic define discuss function vector computation describe generate protein high taxonomy node similar resource good community complex representation demonstrate single compare include 23 obtain optimal 23 24 24 25 Fig. 2. Six topics of ontology alignment based on the bibliometric data 2001-2018. The topics are identified by LDA [5] and visualized in 25 MATLAB. 26 26 27 3. Topic Analysis of Ontology Alignment tion transforms the set of documents into the set 27 28 of bags of words, which are useful for further 28 29 Topic modeling is a statistical approach to discover topic analysis. 29 30 the underlying topics of a set of documents based on – Meaningless Word Removal For meaningful 30 31 the frequency of words that appeared in the documents. analysis, stop words and the words whose length 31 32 Topic modeling is generally used to find the underlying is less than three were removed since they do not 32 33 hidden semantic structure of a text body. One of the have any further impact on the analysis. At the 33 34 most well-known algorithms for modeling the topic is same time, we removed the terms used for the 34 35 latent Dirichlet allocation (LDA) [5], which can find query since they exist in all the retrieved docu- 35 36 the hidden topics of a set of texts with a given number ments, while they do not convey any useful infor- 36 37 of topics. mation for ontology alignment topics. 37 38 In this experiment, we aim at analyzing different – Lemmatization Words were also lemmatized, 38 39 topics in order to provide a broader picture of do- which means that the verbs in the past or fu- 39 40 40 mains and problems to which the ontology alignment ture tense are changed into their present tense 41 41 is an essential contribution. In this regard, the title, and other third person words are replaced by their 42 42 abstract, and keywords of the obtained research out- corresponding first person. We further used the 43 43 puts were subjected to LDA, and the identified topics Porter stemming algorithm [82], which replaces 44 44 were visualized using word clouds. It is typically nec- the words with their associated roots. 45 45 essary to conduct several pre-processing procedures 46 The processed documents were then subjected to 46 on the given data in order for the final topics to be 47 LDA for topic modeling. LDA also needs to have the 47 more meaningful. The following pre-processing strate- 48 number of topics to be identified. We tried a variety of 48 gies are used before applying LDA: 49 numbers between four up until 14, and we found out 49 50 – Tokenization: Tokenization means that sentences that more meaningful topics are obtained by six topics. 50 51 are broken into their constituent words. Tokeniza- Figure 2 displays the topics identified by LDA. In this 51 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative 7

1 figure, the size of each term indicates its importance, ment", “software" (as in software agent), “multi- 1 2 and the most vital terms are also identified by the or- agent", that are visible in Topic 1 of Figure 2. 2 3 ange color. Note that the order of topics is arbitrary, – Web Service Discovery: Web Services contain 3 4 and hence, the topics that appeared at the beginning do the services provided by some providers who ex- 4 5 not have any privilege over those presented later on. pose their particular services to a broad audience 5 6 In the following, we analyze the topics discovered by by using Web technologies. Semantic Web Ser- 6 7 LDA. vices (SWS) are strong tools to describe the ser- 7 8 vices more richly so that their discovery by re- 8 – Agent-based Modeling Agent-based modeling 9 questers become even easier. Web Service discov- 9 is used to simulate the actions and interactions 10 ery is the process of finding a service which is 10 of independent agents in order to estimate their 11 able to deliver a particular service to the person 11 impacts on the overall system. Agents commu- 12 who has requested it. Sometimes a request cannot 12 nicate together by exchanging messages com- 13 be responded merely by a single service, but by a 13 posed in different languages such as Agent Com- 14 number of services. In this circumstance, it is re- 14 15 munication Language developed in Foundation quired to have a composition process, which in- 15 16 for Intelligent Physical Agents (FIPA-ACL) and tegrates several services in order to meet a partic- 16 17 Knowledge Query and Manipulation Language ular need of requesters. SWS can be modeled by 17 18 (KQML). These languages determine only the different standards such as Web Services Descrip- 18 19 overall structure of the messages and not their tion Language (WSDL) [1] and OWL-S [63], and 19 20 contents. The actual content of a message ex- different terminologies might be used by different 20 21 pressed by an agent is typically modeled by an providers/requesters. As a result, the discovery of 21 22 ontology. As a result, when two independent services requires the use of ontology alignment 22 23 agents communicate with each other, it is unlikely techniques so that the discrepancy between dif- 23 24 that they can understand each other if they do ferent services are reconciled and the rate of dis- 24 25 not use the same ontology for communication. In covery success increases significantly. SWS dis- 25 26 this regard, ontology alignment has been exten- covery was also one of the first applications that 26 27 sively used in order for the agents to understand ontology alignment could address. The first paper 27 28 various messages in different formats. One of the employing ontology alignment for SWS discov- 28 29 first problems to which ontology alignment was ery dates back to 2003 [9], and it has been since 29 30 applied is agent communication, where the pa- used in other studies [28, 87, 91]. 30 31 per was published in 2002 [98]. There are also Topic 2 in Figure 2 is devoted to this important 31 32 several systems that employ agent-based mod- application of the ontology alignment. As is read- 32 33 eling for automatic ontology alignment, where ily observable, the terms “web" and “services" are 33 34 they call such systems as agent-based ontology detected as the main terms, and there are also the 34 35 alignment [25, 48, 57]. In this view, the map- terms “discovery" and “composition" as well that 35 36 pings between two ontologies are deemed as a are the general terms in the Semantic Web Service 36 37 product of communications between two intelli- domain. 37 38 gent agents [86]. Topic 1 in Figure 2 illustrates – Process Model Matching: Process models com- 38 39 the topic related to articles of ontology alignment prise a set of related activities or tasks which need 39 40 that used the notions of agent-based modeling. In to be done in a specific order to finally produce 40 41 this topic, as expected, the term “agent" is iden- a service or product. The matching of these pro- 41 42 tified as the most important word. There are also cesses is of the essence for several tasks such 42 43 several other terms related to agent-based mod- as system validation and process harmonization 43 44 eling. For instance, “communication" and “inter- [4, 33, 97]. In this regard, ontology alignment sys- 44 45 action" as are usually paired with “agent", and tems or techniques can be used. A more specific 45 46 the terms “collaborative", “negotiation", and “ex- application is matching the business processes 46 47 change" that refer to collaborations, negotiation, [14, 29, 41, 47], where the process models are 47 48 and data exchange between agents, are the fea- typically related to e-commerce [90]. 48 49 tures that agents can be equipped with by using In the OAEI 2016 and 2017, there was a track 49 50 ontology alignment. There are also some general about matching different process models of the 50 51 terms of agent-based modeling such as “environ- university admission systems. As a result, the 51 8 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative

1 problem is completely well-known for the on- repair the alignment automatically obtained from 1 2 tology alignment community as well. As Topic the alignment systems [64, 65, 78]. 2 3 3 in Figure 2 shows, LDA has been able to de- – Machine Learning and Biomedical Ontology: 3 4 tect the importance of this problem for ontology Topic 6 is a mixture of a well-known approach 4 5 alignment. The terms “process" and “business" for ontology alignment and one essential domain 5 6 are at the heart of this topic, which accentuates to which ontology alignment has been applied. 6 7 the importance of process model matching in on- There are several ontology alignment systems 7 8 tology alignment. The term “management" also which use machine learning techniques for align- 8 9 has a significant weight. Interestingly, “business ment. In fact, machine learning techniques are 9 10 process management" refers to a domain where one of the first approaches that are used for align- 10 11 matching of processes has become a major re- ing ontologies and dates back to 2002 [18, 19]. 11 12 search area [54]. There are also many machine learning-based sys- 12 13 – Query Answering: Information provided by tems that require to have a gold standard for train- 13 14 different sources is not described by a unified ing [40, 62, 84]. These systems are sometimes 14 15 schema on the Web. At the same time, users do called pre-trained systems [26] and need to have 15 16 not utilize the same terminology in their search [a part of] the reference alignment for training. 16 17 queries. Thus, a semantic query answering is re- The system is then ready to map the rest of the 17 18 quired to rewrite the query in order to provide ontologies or other ontologies in the same do- 18 19 sensible results. Since both the information on the main. The terms “learn", “machine", “learning", 19 20 web and the queries are the reasons for the dis- and “classification" in Topic 6 are the indicators 20 21 21 crepancy, ontology alignment can be helpful to of these alignment systems. 22 22 address this challenge and to improve the rele- Another term in this topic is “biomedical", which 23 23 vance of the retrieved information. Thus, ontol- is one of the most important domains to which 24 24 ogy alignment has been used extensively in this the ontology alignment has been applied. The 25 25 regard. As Topic 4 in Figure 2 illustrates, this anatomy track, which is about matching the adult 26 26 problem is quite important in ontology alignment. mouse anatomy and a part of NCI thesaurus com- 27 27 The term “query" is the most accented terms, prising the human anatomy, is one of the first 28 28 and there are some other related terms such as tracks of the OAEI [20]. There are several recent 29 tracks such as disease and phenotype [37] and 29 30 “relational", “database", “schema". In the OAEI 30 2014 and 2015, there was a track for answering large biomedical [43, 44, 46] tracks which have 31 been recently added to the OAEI tracks. There- 31 32 queries by the aid of ontology alignment systems, 32 which indicates that this problem is also quite fore, there is no surprise to see this term as a ma- 33 jor topic of ontology alignment. The terms “large" 33 34 well-known to the community. 34 – Linked Data and Logic: One of the primary ob- and “background" are also related to this theme 35 since there is one large biomedical track in the 35 jectives of the Semantic Web is to link the data on 36 OAEI and it is the common practice to use of 36 the Web to other available resources so that useful 37 background knowledge such as UMLS [7] for 37 information can be provided from the available 38 matching ontologies in the biomedical domain. 38 39 data. Since the published data on the Web are de- 39 40 signed by many people, interlinking of these data 40 41 is not straightforward due to their heterogeneous 41 42 nature. As a result, ontology alignment is a po- 4. Thematic Analysis 42 43 tential solution to fulfill this essential objective of 43 44 the Semantic Web [42, 53]. This is the reason that In this section, the thematic analysis of the ontology 44 45 “linked" in Topic 5 of Figure 2 has been central- alignment publications is presented based on the col- 45 46 ized. Another vital term in this topic is “logic". lected bibliometric data between 2001 and 2018. We 46 47 Logic has been widely used to align two differ- first study the number and types of the research out- 47 48 ent ontologies [45, 55, 71, 72]. One of the well- puts, and it is then followed by the contributions of 48 49 known systems is LogMap [45], which is based ontology alignment publications to the top-cited and 49 50 on logic and is one of the top systems at the OAEI top journal percentiles. Afterward, the disciplines con- 50 51 in the recent decade. Also, logic has been used to tributed to ontology alignment are discussed. 51 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative 9

1 350 1 2 2 3 3 4 300 4 5 5 6 6 250 7 7 8 8 9 9 200 10 10 11 11 12 12 150 13 13 14 14 15 100 15 16 16 17 17 18 50 18 19 19 20 20 21 0 21 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 22 22 23 23 Fig. 3. The number of documents published about ontology alignment on Scopus between 2001 to 2018. 24 24 25 25 4.1. Number and Types of Published Documents research outputs. Interestingly, in 2013, Shvaiko and 26 26 Euzenat [89] showed the improvement of the field 27 27 In this subsection, we discuss the number and the 28 based on their analysis on the state-of-the-art ontology 28 types of research outputs in the ontology matching data 29 matching systems and the results of evaluations, while 29 from 2001 to 2018. The essence of having the auto- 30 they observed that the speed of the ontology alignment 30 matic mapping between two ontologies was discussed 31 progress was slowing down. The slow progress in the 31 for in the late 90s and early 2000 for different prob- 32 field has shown itself in the number of publications in 32 lems such as ontology merging [74, 75] and further in 33 the field as one important criterion. 33 business-to-business (B2B) electronic commerce [76], 34 We also analyze the types of research outputs in the 34 35 where the mappings between ontologies, taxonomies, 35 ontology alignment field. Figure 4 displays the per- 36 and the classification system are required. In the pre- 36 centage of different types of papers published in the 37 ceding years, the existence and importance of such 37 38 an automatic mapping were discussed in several other ontology matching domain between 2001 up until and 38 39 problems such as agent communication in multi-agent including 2018. According to this figure, the vast ma- 39 40 systems [98]. Ever since, ontology alignment has been jority of research outputs, i.e., around 65%, are pub- 40 41 the topics of numerous research studies, by which vari- lished in conference venues. It is no surprise since 41 42 ous problems have been addressed. Figure 3 shows the the main venue for this field is the ontology matching 42 43 number of research outputs from 2001 to 2018. Ac- workshop held in international semantic web confer- 43 cording to this figure, the number of outputs has been 44 ence (ISWC), where there has been several papers and 44 45 steadily increased until 2008, when around 290 re- 45 posters along with the alignment contests. Aside from 46 search articles are published. From 2008 up until 2013, 46 conference papers, journal articles comprise 25% of 47 the number of publications has been approximately 47 48 the same, where the maximum number of outputs is the publications and are ranked as the following types 48 49 in 2013 with 300 publications. From 2013, the num- of publication in ontology matching. Conference re- 49 50 ber of documents has experienced a steady decrease, views and book chapters are the other major types of 50 51 where its minimum number reached in 2018 with 175 articles in this domain. 51 10 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative

1 Conference Paper 1 2 2% 1% Article 2 Conference Review 3 3 Book Chapter 4 7% Others 4 5 5 6 6 7 7 8 8 25% 9 9 10 10 11 11 12 12 13 13 65% 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 Fig. 4. The types of documents published about ontology alignment on Scopus between 2001 to 2018. 21 22 22 23 4.2. Outputs in Top Percentiles Worldwide of the number of research outputs in the top 1% and 23 24 the top 10% most cited articles. 24 25 In this subsection, the appearance of ontology align- We also analyze the share of ontology alignment 25 26 ment research outputs in top-cited and journal per- in the top journals by CiteScore. Figure 6 illustrates 26 27 centiles in the recent six years is explored. the ratio of ontology alignment outputs in top journals 27 28 We first look into the ratio of ontology alignment from 2013 to 2018. According to this figure, the on- 28 29 publications in the top-cited percentiles. In recent six tology alignment share in the top 1% journals is 2.1%, 29 30 years, there are 60 publications in the top 10% most 1.7%, 1.9%, and 1.5% for 2013, 2015, 2017, and 2018, 30 31 cited publications worldwide. The distribution of these respectively. Similarly, the share in the top 10% jour- 31 32 top-cited articles in the recent six years is shown in nals is 13.8% as the highest, followed by 11.8% in 32 33 Figure 5, where the light color shows the percentile 2014, 9.9% in 2015, and 9.2% in 2018. There are a 33 34 in top 10% most cited and the dark color denotes the steady decrease and increase in publishing in the top 34 35 percentile in top 1% most cited articles worldwide. 10% journals from 2013 to 2016 and from 2016 to 35 36 According to this figure, ontology alignment research 2018, respectively. 36 37 outputs constitute one percent, 0.9%, and 0.5% of the 37 38 top 1% most cited articles worldwide in years 2013, 4.3. Fundamental Disciplines to Ontology Alignment 38 39 2015, and 2017, respectively. At the same time, there 39 40 is no ontology alignment output in the top 1% most In this section, we consider the disciplines that con- 40 41 cited for 2014 and 2016. It is also readily seen that on- tribute to ontology alignment. In this regard, we take 41 42 tology alignment outputs form 6.1%, 5.3%, and 5.8% advantage of all science journal classification (ASJC) 42 43 of the top 10% most cited article worldwide in years used in Scopus and visualize the main areas along with 43 44 2013, 2016, and 2018. Interestingly, although the out- their subcategories that contribute to the growth and 44 45 puts from 2016 and 2018, which have a considerable evolution of the ontology alignment field. 45 46 amount of papers in the top 10% most cited articles, Figure 7-(a) displays the main subject areas con- 46 47 they do not have any in the top 1% most cited research tributed to ontology alignment based on the publica- 47 48 outputs worldwide. The top-cited articles in the recent tion data between 2001 and 2018. As expected, com- 48 49 six years are tabulated in Table 2. As expected, four of puter science is the area with the maximum num- 49 50 these articles are published in 2013 so that this year is ber of publications and constitutes 55% of the over- 50 51 considered the best year in the recent six years in terms all research articles. More in details, Figure 7-(b) dis- 51 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative 11

1 8 1 2 2

3 7 3 4 4

5 6 5 6 6

7 5 7 8 8 9 9 4 10 10 11 11 3 12 12 13 13 2 14 14 15 15 1 16 16 17 17 0 18 18 2013 2014 2015 2016 2017 2018 19 % of publication in top 1% most cited % of publication in top 10% most cited 19 20 20 21 Fig. 5. The share of ontology alignment research outputs to the top 1% and the top 10% most cited articles published in all disciplines. 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 18 31 31 32 32 16 33 33 34 34 14 35 35

36 12 36 37 37

38 10 38 39 39 40 8 40 41 41 42 6 42 43 43 4 44 44 45 45 2 46 46 47 47 0 48 2013 2014 2015 2016 2017 2018 48 49 % of publications in top 1% journals % of publications in top 10% journals 49 50 50 51 Fig. 6. The share of ontology alignment research outputs to the top 1% and the top 10% journals of all disciplines. 51 12 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative

1 Table 2 1 2 Five top-cited publications in ontology alignment in the recent six years. 2 3 Title Main Authors Year 3 4 1 Ontology matching: State of the art and future challenges Shvaiko, P., Euzenat, J. 2013 4 5 2 Ontology matching: Second edition Shvaiko, P., Euzenat, J. 2013 5 6 3 Ontology matching: A literature review Otero-Cerdeira, L., RodrÃ guez-MartÃ nez, F.J. 2015 6 7 4 Scaling semantic parsers with on-the-fly ontology matching Kwiatkowski, T., Choi, E., Artzi, Y. 2013 7 8 8 5 The AgreementMakerLight ontology matching system Faria, D., Pesquita, C., Santos, E. 2013 9 9 10 10 11 plays the subcategories of computer science contribut- cording to this figure, health informatics (70.5%), gen- 11 12 ing to ontology alignment. According to this figure, eral medicine (8.2%), and medicine (miscellaneous) 12 13 general computer science has the most of published (8.2%) are the subcategories with contributions to on- 13 14 articles, followed by software (12.2%), computer net- tology alignment. 14 15 works and communication (11.8%), and information 15 16 systems (9.7%). 16 17 The second major discipline to the ontology align- 5. Research Collaboration in Ontology Alignment 17 18 ment development is mathematics, which forms 14.8% 18 19 of the overall research outputs. In particular, Figure 7- The collaboration of researchers within an area 19 20 (c) illustrates the subcategories of mathematics, which widens the impact and scope of the corresponding sci- 20 21 indicates that theoretical computer science makes up entific field. As a result, it is of the essence to mon- 21 22 60.9% of the overall articles related to this category, itor, and even encourage, the collaborations between 22 23 and it is followed by general mathematics (9.3%) and different researchers and institutes all over the world. 23 24 modeling and simulation (9.0%). In this section, the research collaboration in ontology 24 25 The third main area is engineering, which forms alignment is investigated. In this regard, we consider 25 26 10.8% of all publications in this field. More in details, the collaboration between authors and countries and 26 27 Figure 7-(d) displays the subcategories of engineer- identify their most collaborative elements. We further 27 28 ing contributing to ontology alignment. According to analyze the trend of collaborations in recent years and 28 29 this figure, control and system engineering has 29.7% academic-corporate collaborations in ontology align- 29 30 of publications and is followed by general engineer- ment. 30 31 ing (26.4%) and electrical and electronic engineering 31 32 (20.4%). 5.1. Author Collaboration 32 33 Social science is another important area contributing 33 34 to ontology alignment, which constitutes 3.8% of all The collaborations among authors of ontology align- 34 35 publications. Figure 7-(e) shows the subcategories of ment are visualized in Figure 8 based on the bibliomet- 35 36 social science which have the most share in ontology ric data 2001-2018. The authors in this graph are rep- 36 37 alignment research outputs. According to this figure, resented by nodes, where their size is proportionate to 37 38 library and information system has the largest part of the number of collaborative articles, and their color de- 38 39 publications, i.e., 39.2%, and education (19.6%) and notes the average number of citations per publication 39 40 linguistic and language (16.5%) follow it. according to the collected data from 2001-2018. Also, 40 41 Decision science has 3.3% of overall research out- the thickness of each edge between two authors is com- 41 42 puts in ontology alignment. Figure 7-(f) shows that in- mensurate with the number of collaborative publica- 42 43 formation system and management (83.3%), general tions of the authors at the two ends. 43 44 decision science (11.1%), and management science According to Figure 8, it is evident that the orga- 44 45 and operations research (5.6%) are the subcategories nizers of the OAEI have great partnerships, and the 45 46 in this category with contributions to the development most collaborative authors are also coming from this 46 47 of ontology alignment. community. In particular, J. Euzenat, P. Shvaiko, and 47 48 As the last subject area contributing to ontology E. Jiminez-Ruiz have 82, 63, and 63 co-publications, 48 49 alignment, medicine makes up 2.5% of publications. respectively, and lead the list of the author collabora- 49 50 More in details, Figure 7-(g) displays the subcategories tions. Table 3 tabulates the top collaborative authors 50 51 of this area with the shares in research outputs. Ac- in ontology alignment. In this table, the number of ar- 51 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative 13

1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 (a) The main subject areas of ontology alignment. 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 (b) Subcategories of computer science (c) Subcategories of mathematics 28 29 29 30 30 31 31 32 32 33 33 34 34 35 35 36 36 37 (d) Subcategories of engineering (e) Subcategories of social science 37 38 38 39 39 40 40 41 41 42 42 43 43 44 44 45 45 46 46 47 47 48 (f) Subcategories of decision science (g) Subcategories of medicine 48 49 49 50 Fig. 7. The Disciplines and their associated subcategories contributed to ontology alignment. 50 51 51 14 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative

1 ticles co-authored with others, the total number of all is a signal of field evolution from a solitary enterprise 1 2 co-authors, and the average citation per paper are dis- to an expanding social movement [39]. 2 3 played in the first to third columns, respectively. Ac- To further detect the communities of collaborations 3 4 cording to this table, J. Euzanat, P. Shvaiko, and C. in ontology alignment, the co-authorship network in 4 5 Trojahn have the maximum number of collaborations Figure 8 was subjected to a community detection algo- 5 6 in terms of the total number of co-authors. rithm, and the major collaborative communities were 6 7 Aside from the OAEI organizers, the developers of detected accordingly. There are quite a few commu- 7 8 AgreementMaker and AgreementMakerLight (AML) nity detection algorithms [30], and we choose Lou- 8 9 [27], i.e., D. Faria, C. Pesquita, and I. Cruz, that are vain algorithm [6], due to its speed, scalability, and 9 10 positioned at the top of Figure 8, have significant col- simplicity [11, 24]. It is also one of the most popular 10 11 laborations with each other. Also, D. Faria is one of community detection algorithms and has been imple- 11 12 the OAEI organizers, and along with C. Pesquita, have mented in many software and programming packages 12 13 several collaborations with the OAEI community as such as Gephi. Figure 9 plots the identified communi- 13 14 well. Other members in this cluster have partnerships ties, where each community is depicted in a particu- 14 15 solely with each other. lar color. Six communities were identified by the algo- 15 16 Another dense cluster is placed at the bottom left rithm, two of which are quite significant and include 16 17 of Figure 8. A closer look at the authors indicates that 75% of all researchers in this domain, and are shown 17 18 they are mainly from China, and have few collabora- in yellow and purple colors. The former is the OAEI 18 19 tions with researchers outside of their country. The ex- organizers who have created a very large community. 19 20 ceptions are S. Wang with 46, Juanzi Li with 48, and The thickness of edges of this network also indicates 20 21 21 Yingjie Li with 40 co-authorship, including collabo- that researchers in the OAEI community greatly col- 22 laborate with each other. The other dominant commu- 22 ration with other groups around the world. The main 23 nity is the Chinese, in which the collaborations, though 23 reason of such collaborations is that they are mainly 24 not as significant as the OAEI community, are remark- 24 studied outside of China, and had a better opportunity 25 able. Also, this figure indicates that the collaborations 25 for international collaborations. Other authors from 26 between Chinese and OAEI communities are not of 26 this community who had mainly collaborated intra- 27 significance, and researchers primarily cooperate with 27 nationally are Y. Wang with 44, J. Wang with 37, and 28 other researchers from the same community. 28 S. Zhang with 32 co-authorships. 29 29 The color of nodes in Figure 8 is proportionate to 30 5.2. Country Collaboration 30 31 the average number of citations, where nodes closer 31 to the yellow color have a higher number of citations 32 In this section, the collaborations between coun- 32 per paper. It is interesting to see that the OAEI orga- 33 tries are discussed and visualized. Figure 10 displays 33 34 nizers, who vastly collaborate with other researchers, the co-authorship between different countries using a 34 35 have the maximum citation average compared to oth- graph. The nodes of this graph are the countries with 35 36 ers. Based on Table 3, Shvaiko with 116.58 citations the size of node being proportionate to the number 36 37 per publication leads the author list in terms of aver- of publications collaborating with other countries, and 37 38 age citations, followed by J. Euzenat with 70.92, and the strength of the edge between each pair of countries 38 39 A. Ferrara with 31.76 citations per paper. Another im- is commensurate to the number of publications written 39 40 portant point is about the Chinese community. Among jointly by authors of the corresponding countries. Ac- 40 41 these researchers, the authors who have collaborations cording to this figure, France, UK, Germany, the US, 41 42 with international communities have higher average ci- and Italy are respectively the top five countries having 42 43 tations. This confirms previous studies that multina- the most collaborative papers worldwide. 43 44 tional research collaboration are associated with in- The top five countries have the most collaborative 44 45 creased citations [39, 96]. In general, international re- research outputs together. For French authors of ontol- 45 46 search collaboration is recognized as a means of culti- ogy alignment publications, the international collabo- 46 47 vating research quality, enhanced resource utilization, ration is mostly with Germany (with 13% share of all 47 48 and high impact [39, 83]. It also has indirect strategic, collaborative research outputs), Italy (12%) and the US 48 49 economic, or political benefits [34]. In fact enlarging (8%). For the UK, the co-authorship is most frequently 49 50 team sizes, increasing interdisciplinarity, and intensi- with Italy (14%), Germany 13%, and the US (11%). 50 51 fying ties across institutional and geographic borders German international co-authorships are mostly dom- 51 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative 15

1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 Fig. 8. Collaborations of authors in ontology alignment based on the bibliometric data 2001-2018. The size of nodes is proportionate to the 25 26 number of collaborative publications by the associated author, the color of nodes represent the average citations of authors per paper, and the 26 27 thickness of each edge is commensurate with the number of collaborative articles of the authors at the two ends. 27 28 28 Table 3 29 29 The ontology alignment researchers with the maximum number of collaborative publications. The ﬁrst column tabulates the number of collabo- 30 30 rative articles, the second column denotes the number of all collaborative researchers, and the third column is the average citation per publication. 31 31 Author # Co-authored Articles # All Co-authors Average Citations 32 32 33 Euzenat J. 82 240 70.92 33 34 Shvaiko P. 63 205 116.58 34 35 Jiminez-Ruiz E. 63 168 16.86 35 36 Trojahn C. 59 204 13.98 36 37 StuckenSchmidt H. 59 177 18.95 37 38 Ferrara A. 56 163 31.76 38 39 Meilicke C. 54 160 22.72 39 40 40 41 inated by collaborations with France (14%), the UK We further analyze the collaboration between coun- 41 42 (13%), and Italy (12%). US international ontology tries along with the number of their publications and 42 43 alignment collaborations are also most commonly with citations they have received. In this regard, Figure 11 43 44 the UK (12%), then with researchers from France plots the collaboration between countries with the size 44 of nodes being proportional to the number of the publi- 45 (10%) and Italy (9%). Italian ontology researchers 45 46 cations of the country, while Figure 12 shows the same 46 most frequently collaborate with researchers from the 47 graph with the size of nodes being commensurate to 47 UK (16%), then with colleagues from France (14.3%) 48 the average number of citations per publication. Ac- 48 49 and Germany (14%). In sum, these ﬁve countries are cording to Figure 11, China has the maximum publi- 49 50 the most collaborative countries worldwide that mainly cations among other countries with 578 research out- 50 51 cooperate with each other. puts forming around 20% of all ontology alignment re- 51 16 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative

1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 34 Fig. 9. Communities of collaborations in ontology alignment based on the bibliometric data 2001-2018. 34 35 35 36 search outputs. At the same time, the number of cita- dom, which is then emerged as more publications for 36 37 tions of this country is not commensurate to the num- the whole group. 37 38 ber of publications. That is to say, the research out- 38 39 puts with at least one Chinese author have not gained 5.3. International Collaboration 39 40 enough attention. One of the primary reasons is the 40 41 lack of international collaborations, as Figure 10 sim- In this subsection, we take a closer look at the lev- 41 42 ply explains. The top ﬁve collaborative countries are els of collaborations in the recent six years. In this re- 42 43 readily seen to have more publications and citations gard, we count the number of research outputs pub- 43 44 among other countries. This corroborates the impor- lished by authors from different countries (interna- 44 45 tance of collaborations since collaborations increase tional collaboration), people from different institutes 45 46 the visibility of research outputs to a wide audience, within a country (only national collaboration), differ- 46 47 and the research studies thus get the attention they de- ent researchers of an institute (only institutional col- 47 48 serve. Another important point is the positive correla- laboration), and single-authored ones (no collabora- 48 49 tion between co-authorships and the size of publication). Figure 13 plots the share of ontology alignment 49 50 tion outputs. It also makes sense since cooperation be- published articles in each of these categories. Accord- 50 51 tween researchers helps the use of the common wis- ing to this ﬁgure, as expected, the maximum collabo- 51 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative 17

1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 Fig. 10. Collaborations of countries in ontology alignment based on the bibliometric data 2001-2018. The size of nodes represents the number 34 of all collaborative papers of researchers from the associated country. 34 35 35 36 rations are among the researchers from the same insti- declined, while the single-authored articles have been 36 37 tutes, and the only national and only institutional col- growing. Ironically, the share of international collabo- 37 38 laborations have approximately similar shares of on- rations (the most desired) and the share of no collab- 38 39 tology alignment research outputs. Single-authored ar- oration have been increased together over the last few 39 40 ticles are at the bottom of this list and from around years. 40 41 41 10% of overall publications. 42 5.4. Academic-Corporate Collaboration 42 It is also seen from Figure 13 that international col- 43 43 laboration has increased in the recent six years, from 44 In this section, we consider the relationship be- 44 18.9% in 2013 to 27% in 2018. The maximum in- 45 tween academic institutions and corporates based on 45 46 ternational collaborations are also in the year 2017, ontology alignment published articles in the recent six 46 47 which constitute 28.9% of all ontology alignment re- years. 47 48 search outputs. The collaborations inside the institu- In an academic-corporate relationship, collaboration 48 49 tions have experienced a signiﬁcant decrease, from is of the essence for both sides. It helps the academia to 49 50 54.6% in 2013 to 45.5% in 2018. Similarly, the col- ensure industrial relevance in its research [99], and on 50 51 laborations between institutions within a country have the other hand, it provides the opportunity of knowl- 51 18 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative

1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 Fig. 11. Collaborations of countries in ontology alignment based on the bibliometric data 2001-2018. The size of nodes is proportionate to the 32 33 number of published documents with at least one author from the associated country. 33 34 34 35 edge complementary and risk sharing with the corpo- 6. Contribution and Impact in Ontology 35 Alignment 36 rates [2]. Figure 14 displays the share of published ar- 36 37 37 38 ticles by the academic-corporate collaborations. Ac- In this section, the impact of authors and countries 38 39 cording to this figure, a tiny portion of papers is pub- that are active in ontology alignment is investigated. In 39 40 this regard, we count the number of documents pub- 40 41 lished jointly by academia and corporates, with a max- lished per authors or countries and the number of cita- 41 42 imum of 3% in 2014. The minimum portion is also tions of their documents to analyze their influence on 42 43 ontology alignment. 43 44 from the years 2013 and 2018 with 1.7% of all on- 44 45 tology alignment published articles in the correspond- 6.1. Contribution and Impact of Authors in Ontology 45 46 Alignment 46 47 ing years. This figure elaborates that the relation be- 47 48 tween industry and academic institutions is not strong In this subsection, the influence of authors on on- 48 49 tology alignment is investigated by counting the pub- 49 enough sine vast majority of collaborations is inside 50 lications of top authors and their number of citations. 50 51 either academia or industry. Figure 15 plots the publication count of the top 10 au- 51 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative 19

1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 Fig. 12. Collaborations of countries in ontology alignment based on the bibliometric data 2001-2018. The size of nodes is proportionate to the 32 33 number of citations of the corresponding country. 33 34 34 35 thors in ontology alignment. The top 10 authors have the author using VOSviewer. Figure 17 visualizes the 35 36 published around 308 articles spanning from 2001 to citation map of ontology alignment authors, where the 36 37 2018, which constitute 10.3% of all publications in on- size of nodes and their colors are proportionate to the 37 38 tology alignment. According to Figure 15, J. Euzenat citations and average citations per paper, respectively. 38 39 tops the list of authors with 70 publications in ontology Also, the thickness of edges is proportional to the num- 39 40 alignment, and C. Trojahn with 50, E. Jimenez-Ruiz ber of times the authors at the two ends have cited each 40 41 and H. Stuckenschmidt with 45 and P. Shvaiko with 38 other. Aside from the top authors shown in Figure 16, 41 42 research outputs follow. another critical observation from Figure 17 is that the 42 43 We further analyze the ontology alignment authors researchers who are active in the OAEI have received 43 44 in terms of their number of citations. In this regard, more attention in comparison to others. The bottom- 44 45 Figure 16 plots the top 10 authors with the maximum right region of this ﬁgure shows that the OAEI organiz- 45 46 number of citations. According to this ﬁgure, J. Eu- ers and participants have higher average citations than 46 47 zenat leads the list with 4,610 citations, and P. Shvaiko other researchers. 47 48 with 3,847, M. Scholemmer with 1,040, and Stucken- Aside from the OAEI organizers and participants, 48 49 schmidt with 834 citations follow. several other researchers have been cited well. Y. 49 50 To provide a broader view of the author impacts on Kalfoglou and M. Schorlemmer have a well-cited re- 50 51 ontology alignment, we visualize the citation map of view paper published in 2003, and two methods for 51 20 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative

1 60 1 2 2 3 3 50 4 4 5 5

6 40 6 7 7 8 8 9 30 9 10 10 11 11 20 12 12 13 13 14 10 14 15 15 16 16 17 0 17 2013 2014 2015 2016 2017 2018 18 18 International Collaboration Only National Collaboration 19 19 Only Institutional Collaboration Single Authorship (No Collaboration) 20 20 21 21 Fig. 13. The share of different types of Collaborations in ontology alignment in the recent six years. 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 3.5 31 31 32 32 33 3 33 34 34 35 35 2.5 36 36 37 37

38 2 38 39 39 40 40 1.5 41 41 42 42

43 1 43 44 44 45 45 46 0.5 46 47 47 48 48 0 49 2013 2014 2015 2016 2017 2018 49 50 50 51 Fig. 14. The share of academic-corporate collaborations in ontology alignment. 51 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative 21

1 1 2 Euzenat, J. 2 3 3 4 Trojahn, C. 4 5 5 6 Xue, X. 6 7 7 8 Stuckenschmidt, H. 8 9 9 10 Jiménez-Ruiz, E. 10 11 11 12 Shvaiko, P. 12 13 13 14 Meilicke, C. 14 15 15 16 Gao, W. 16 17 17 18 Pesquita, C. 18 19 19 20 Lambrix, P. 20 21 21 0 10 20 30 40 50 60 70 80 22 22 23 23 Fig. 15. The top 10 authors of ontology alignment in terms of their number of published documents based on the bibliometric data 2001-2018. 24 24 25 25 26 26 27 27 28 28 29 29 Euzenat, J. 30 30 31 31 Shvaiko, P. 32 32 33 33 34 Schorlemmer, M. 34 35 35 36 Kalfoglou, Y. 36 37 37 38 Stuckenschmidt, H. 38 39 39 40 40 Rahm, E. 41 41 42 42 Meilicke, C. 43 43 44 44 45 Ferrara, A. 45 46 46 47 Noy, N. 47 48 48 49 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 49 50 50 51 Fig. 16. The top 10 authors of ontology alignment in terms of their number of citations based on the bibliometric data 2001-2018. 51 22 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative

1 ontology alignment published in 2002 and 2003 [49– tions: While they publish more than any country, they 1 2 51]. N. Noy has also conducted several fundamen- are ranked sixth concerning the number of citations. 2 3 tal research in ontology alignment in the first years Italy has the highest average citations per publica- 3 4 of this century [66, 73–75]. Y. Li, J. Tang, and J. Li tion with 29.54 for 204 documents and is followed by 4 5 have developed (along with colleagues) the ontology France with 21.45, Spain 20.04, the US with 18.51, 5 6 alignment system, called RiMOM, where their semi- Germany with 15.33, and the UK with 14.65 citations 6 7 nal work was published in 2008 [92? ]. M. Ehrig has per publication. Moreover, based on Figure 20, the US 7 8 published the book entitled ontology alignment: bridg- has the maximum number of citations, followed by 8 9 ing the semantic gap in 2006, which has got more than France, Italy, Germany, and the UK. 9 10 600 citations to date [22]. As one can readily realize, In addition, the community algorithm has been ap- 10 11 these researchers have received a great amount of at- plied to the network in Figure 20. A community from 11 12 tention because of their research in the first decade of this figure is a set of countries whose researchers re- 12 13 this century. However, there are several well-cited refer to each others’ studies more often. Basically, two 13 14 searchers among the OAEI organizers, whose research major communities were detected that are shown in 14 15 studies are more recent. The examples are E. Jiminez- purple and green colors. One community comprises 15 16 Ruiz who developed LogMap [45], D. Faria and C. the US, UK, China, and Spain along with some other 16 17 Pesquita, who developed AML [27]. As a result, the countries with less significant contribution to ontol- 17 18 attention of ontology alignment is more focused on the ogy alignment. Another community consists of France, 18 19 OAEI in recent years, and other researchers in this field Italy, and Germany as the major countries along with 19 20 have not drawn significant attention. some less considerable ones such as the Netherlands 20 21 and Portugal. 21 22 6.2. Contribution and Impact of Countries in 22 23 Ontology Alignment 23 24 7. Conclusion and Discussion 24 25 In this section, we analyze the impact and contribu- 25 26 tions of different countries on the evolution of ontol- In this paper, we revisited ontology alignment by 26 27 ogy alignment. First, we visualize the number of on- using bibliometric analyses. In this regard, we made 27 28 tology alignment publications for each country. Figure an inquiry in Scopus and extracted articles pertinent 28 29 18 displays the top 10 countries in terms of their publi- to ontology alignment. After inspecting the articles 29 30 cation count. According to this figure, China leads the and excluding irrelevant items to ontology alignment, 30 31 list with 578 publications, which has a quite well dif- 2,975 articles remained based on which the quantita- 31 32 ference with the second country, the US, with 360 pub- tive analyses were carried out. 32 33 lications in ontology alignment. France with 306, Ger- We conducted two classes of analyses, namely, se- 33 34 many with 295, and the UK with 234 publications in mantic and structural. Semantic analysis entails the 34 35 ontology alignment are the following countries in this overall discovery of concepts, notions, and research 35 36 list. lines flowing underneath ontology alignment, while 36 37 We further analyze the countries concerning their the structural analysis provides a meta-level overview 37 38 number of citations. Figure 19 shows the top 10 coun- of the field by probing into the collaboration network 38 39 tries regarding the number of citations. Based on this and citation analysis in author and country levels. Each 39 40 figure, the US tops the list with 6,662 citations, and it is of these analyses was divided into two subcategories. 40 41 followed by France with 6563 and Italy with 6027 ci- In the semantic analysis, we first conducted a topic 41 42 tations. China, while has the maximum number of on- modeling on the extracted bibliometric data by sub- 42 43 tology alignment citations, is the sixth country in this jecting title, keywords, and abstracts of articles to the 43 44 ranking with 3080 citations overall. latent Dirichlet allocation (LDA), a well-established 44 45 Besides, we visualize the citation network of coun- statistical method for modeling topics. Although the 45 46 tries using VOSviewer. Figure 20 illustrates the cita- topics were detected based merely on the frequency 46 47 tion network where the nodes are the countries whose of the words in the articles, the identified topics read- 47 48 size are proportionate to the number of citations based ily referred to a problem that ontology alignment can 48 49 on the bibliometric data 2001-2018. According to this address or a domain to which it has been applied. 49 50 figure, the number of citations of Chinese researchers The other semantic analysis was thematic, wherein the 50 51 is not commensurate with their number of publica- number of annual publications, the types of research 51 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative 23

1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 Fig. 17. The citation map of ontology alignment researchers, where the size of nodes is proportionate with the number of citations and its color 21 22 represents the average citations per paper for the associated researcher. The thickness of edges in this network is proportional to the number of 22 23 times the authors at the ends have cited each other. 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 China 32 32 33 United States 33 34 34 35 France 35 36 36 Germany 37 37 38 38 United Kingdom 39 39

40 Italy 40 41 41 42 Spain 42 43 43 44 Netherlands 44 45 45 Portugal 46 46 47 47 Brazil 48 48 49 0 100 200 300 400 500 600 700 49 50 50 51 Fig. 18. The top 10 countries of ontology alignment in terms of their number of published documents based on the bibliometric data 2001-2018. 51 24 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative

1 1 2 United States 2 3 3 France 4 4 5 5 Italy 6 6 7 Germany 7 8 8 9 United Kingdom 9 10 10 China 11 11 12 12 Spain 13 13

14 Netherlands 14 15 15 16 Greece 16 17 17 18 Portugal 18 19 19 0 1000 2000 3000 4000 5000 6000 7000 20 20

21 Fig. 19. The top 10 authors of ontology alignment in terms of their number of citations based on the bibliometric data 2001-2018. 21 22 22 23 outputs, the share of ontology alignment in top-cited analyses, we identified the authors and countries with 23 24 articles and top journals, and contributing disciplines maximal influence on the field by counting their num- 24 25 to ontology alignment were explained and discussed in ber of publications and citations. We also visualized 25 26 detail. their citation network, where we identified two hosts 26 27 The second class of analyses was structural, wherein of countries that mostly cite each other. 27 28 we carried out two classes of probings. First, we We observed that the articles from the ontology 28 29 29 studied the collaborations of ontology alignment re- matching workshop are not indexed by Web of Sci- 30 30 searchers. We observed that international collaboration ence (WoS). This workshop is particularly important 31 31 has been improved over the last few years. Also, the for the ontology alignment community since highly- 32 32 collaboration between academia and corporate were cited researchers from the field actively participate in 33 33 gauged, which was not significant enough. We also the venue. Also, the articles from the workshop repre- 34 34 observed that ontology alignment researchers fall into sent the state-of-the-art challenges and novelties in this 35 35 36 two major communities. The first community com- domain. In addition, the OAEI contest is also a part of 36 37 prises the OAEI organizers, and the second consists of this workshop, wherein new challenges are introduced 37 38 the Chinese researchers. Although the researchers in to be tackled and, new alignment systems participate 38 39 these communities cooperate closely with each other, to overcome those challenges. 39 40 the communities, especially the Chinese, seem like Based on the contributions of ontology alignment to 40 41 isolated islands that do not interact with the other com- the top-cited articles and top journal percentiles, we 41 42 munity. We also observed that, although the number perceived that ontology alignment is indeed a very es- 42 43 of publications with at least one Chinese researcher is sential field of study, the research outputs of the do- 43 44 considerable, they do not get enough attention, possi- main receive significant attention, and fundamental re- 44 45 bly the attention that they deserve. Aside from authors, search studies are conducted and published in top jour- 45 46 we also analyzed the collaborations between countries nals. However, this field has only one venue, the on- 46 47 and realized that the top five most collaborative coun- tology matching workshop, which, ironically, is not 47 48 tries mostly work together more than any other coun- indexed by WoS. This calls for having more serious 48 49 try in the world. Another structural analysis was re- venues such as a dedicated journal for ontology align- 49 50 garding the impact and contribution of researchers and ment. The pertinent articles to ontology alignment are 50 51 countries for the field of ontology alignment. In these dispersed in various journals, while the conference pa- 51 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative 25

1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 18 18 19 19 20 20 21 21 22 22 23 23 24 24 25 25 26 26 27 27 28 28 29 29 30 30 31 31 32 32 33 33 Fig. 20. The citation map of countries contributing to ontology alignment, where the size of nodes is proportionate to the number of citations. 34 34 35 35 pers are mainly published in the ontology matching ful for the biomedical domain since most of the tracks 36 36 workshop. Having a journal devoted to ontology align- are from this domain, i.e., anatomy, disease and phe- 37 37 ment helps the research in this field to be well-focused notype, and large biomedical. While it is indisputable 38 38 39 in one venue. Also, there have been several special is- that ontology alignment has been successfully applied 39 40 sues on ontology alignment over the last few years that to various biomedical ontologies, its use and applica- 40 41 are also hosted by different journals. A dedicated jour- bility are not restricted to this domain. We encourage 41 42 nal can be solely the host of these special issues. the OAEI organizers and other researchers in the field 42 43 The topics identified in the semantic analysis were to prepare several other standard benchmarks for eval- 43 44 directly related to the problems and applications re- uating ontology alignment systems. The benchmarks 44 45 lated to ontology alignment. Some of these topics are can be from Semantic Web Services, agent-based mod- 45 46 well-established in ontology alignment, and the re- eling, knowledge graphs, and business processes. 46 47 searchers of the field are well aware of it. However, We also identified two primary communities for on- 47 48 there are several other topics that are totally neglected tology alignment, in both of which around 75% of 48 49 by researchers and the OAEI organizers in particular. ontology alignment researchers fall. The collabora- 49 50 By viewing the OAEI benchmarks, one readily gets the tions between these two communities are not signif- 50 51 impression that ontology alignment is particularly use- icant, and researchers of these communities collabo- 51 26 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative

1 rate mainly with other researchers in the same com- for the progress of ontology matching is to dedicate 1 2 munity. Based on the citation analysis, we realized that more research funding to the applicability of ontol- 2 3 the research of the Chinese community does not get ogy alignment and find untapped domains and prob- 3 4 considerable attention, while the number of research lems to which ontology alignment is a potential so- 4 5 outputs of this community is significant. One reason lution. Such projects create new benchmarks for the 5 6 for not getting enough attention is probably the lack OAEI and extend the ontology alignment applicabil- 6 7 of international collaboration in the Chinese commu- ity so that a broader audience can understand the im- 7 8 nity. Another reason might be that the contributions of portance of the field. This also can help overcome the 8 9 this community do not address the current state-of-the- slow progress we have observed from 2013. 9 10 art challenges so that they do not draw enough atten- 10 11 tion. In any case, ontology alignment should take ad- 11 12 vantage of the potentials in the Chinese community. 12 References 13 Thus, the collaborations between these two communi- 13 14 ties are strongly recommended so that the Chinese re- 14 [1] Rama Akkiraju, Joel Farrell, John A Miller, Meenakshi Na- 15 searcher would be familiar with the current challenges 15 garajan, Amit P Sheth, and Kunal Verma. 2005. Web service 16 in the domain, they can disseminate their research to semantics-wsdl-s. (2005). 16 17 a wider audience, and can play a more vital role for [2] Omar Al-Tabbaa and Samuel Ankrah. 2018. 17 18 progressing the field of ontology alignment. A poten- âAŸEngineeredâ˘ A˘ ZUniversity-Industry´ Collaboration: A 18 19 tial direction for future research is to inspect the re- Social Capital Perspective. European Management Review 19 20 search of these two communities to see precisely what (2018). 20 [3] Mathieu Bastian, Sebastien Heymann, and Mathieu Jacomy. 21 they are working on, and then find the potential topics 21 2009. Gephi: an open source software for exploring and ma- 22 22 on which these communities can cooperate with each nipulating networks. In Third international AAAI conference 23 other. on weblogs and social media. 23 24 We discussed and visualized the impact of re- [4] Yacine Belhoul, Mohammed Haddad, Ahmed Gater, Daniela 24 25 searchers and countries on ontology alignment as well. Grigori, Hamamache Kheddouci, and Mokrane Bouzeghoub. 25 26 We observed that the OAEI organizers and participants 2013. Spectral graph approach for process model matchmak- 26 27 get a considerable amount of attention. There are sev- ing. In 2013 IEEE International Conference on Services Com- 27 puting. IEEE, 408–415. 28 eral other researchers with significant attention, but 28 [5] David M Blei, Andrew Y Ng, and Michael I Jordan. 2003. La- 29 29 their research studies were mainly carried out in the tent dirichlet allocation. Journal of machine Learning research 30 first decade of this century. Thus, if researchers want 3, Jan (2003), 993–1022. 30 31 to get attention, they must be involved in the OAEI [6] Vincent D Blondel, Jean-Loup Guillaume, Renaud Lambiotte, 31 32 community and participate actively in the contest. and Etienne Lefebvre. 2008. Fast unfolding of communities in 32 33 Another critical observation from the analyses in large networks. Journal of Statistical Mechanics: Theory and 33 Experiment 2008, 10 (10 2008), P10008. 34 this article is the insignificant collaborations between 34 [7] Olivier Bodenreider. 2004. The unified medical language sys- 35 35 academia and corporate. We observed that the academic- tem (UMLS): integrating biomedical terminology. Nucleic 36 corporate relations constitute on average, around 2% acids research 32, suppl_1 (2004), D267–D270. 36 37 of all publications in ontology alignment. One possible [8] Lindell Bromham, Russell Dinnage, and Xia Hua. 2016. Inter- 37 38 reason is that the teams inside the enterprises have the disciplinary research has consistently lower funding success. 38 39 ability to resolve the problems. However, that does not Nature 534, 7609 (2016), 684. 39 40 explain the inconsiderable relation between academia [9] Mark H Burstein. 2003. The many faces of mapping and trans- 40 lation for semantic web services. In Proceedings of the Fourth 41 and industry. The more realistic reason is that the com- 41 International Conference on Web Information Systems Engi- 42 42 panies have not realized that ontology alignment can neering, 2003. WISE 2003. IEEE, 261–268. 43 enhance their business functions. This is also coming [10] Cristian Candia, C Jara-Figueroa, Carlos Rodriguez-Sickert, 43 44 from the fact that most of the standard benchmarks Albert-László Barabási, and César A Hidalgo. 2019. The uni- 44 45 are restricted to particular problems and domain, while versal decay of collective memory and attention. Nature Hu- 45 46 ontology alignment can be widely used to address dis- man Behaviour 3, 1 (2019), 82. 46 47 parate issues. One way to increase the impact and use [11] Pravin Chopade and Justin Zhan. 2016. A framework for com- 47 munity detection in large networks using game-theoretic mod- 48 of ontology alignment is to conduct several qualitative 48 eling. IEEE Transactions on Big Data 3, 3 (2016), 276–288. 49 49 case studies to show how ontology alignment can au- [12] Caspar Chorus and Ludo Waltman. 2016. A large-scale analy- 50 tomate the manual procedures and consequently, in- sis of impact factor biased journal self-citations. PLoS One 11, 50 51 crease the profits of the companies. Another avenue 8 (2016), e0161021. 51 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative 27

1 [13] Susan Cozzens, Sonia Gatchair, Jongseok Kang, Kyung-Sup [29] Janina Fengel. 2014. Semantic technologies for aligning het- 1 2 Kim, Hyuck Jai Lee, Gonzalo Ordóñez, and Alan Porter. 2010. erogeneous business process models. Business Process Man- 2 3 Emerging technologies: quantitative identification and mea- agement Journal 20, 4 (2014), 549–570. 3 surement. Technology Analysis & Strategic Management 22, 3 [30] Santo Fortunato. 2010. Community detection in graphs. 4 4 (2010), 361–376. Physics Reports 486, 3-5 (2 2010), 75–174. 5 5 [14] Remco Dijkman, Marlon Dumas, Luciano Garcia-Banuelos, [31] Santo Fortunato, Carl T Bergstrom, Katy Börner, James A 6 and Reina Kaarik. 2009. Aligning business process models. In Evans, Dirk Helbing, Staša Milojevic,´ Alexander M Petersen, 6 7 2009 IEEE International Enterprise Distributed Object Com- Filippo Radicchi, Roberta Sinatra, Brian Uzzi, et al. 2018. Sci- 7 8 puting Conference. IEEE, 45–53. ence of science. Science 359, 6379 (2018), eaao0185. 8 9 [15] Ying Ding, Gobinda G Chowdhury, and Schubert Foo. 2001. [32] Morgan R Frank, Dashun Wang, Manuel Cebrian, and Iyad 9 Bibliometric cartography of information retrieval research by 10 Rahwan. 2019. The evolution of citation graphs in artificial in- 10 using co-word analysis. Information processing & manage- telligence research. Nature Machine Intelligence 1, 2 (2019), 11 11 ment 37, 6 (2001), 817–842. 79. 12 [16] Ying Ding and Schubert Foo. 2002. Ontology research and [33] Ahmed Gater, Daniela Grigori, and Mokrane Bouzeghoub. 12 13 development. Part 1-a review of ontology generation. Journal 2010. Complex mapping discovery for semantic process model 13 14 of information science 28, 2 (2002), 123–136. alignment. In Proceedings of the 12th International Confer- 14 [17] Ying Ding and Schubert Foo. 2002. Ontology research and de- 15 ence on Information Integration and Web-based Applications 15 velopment. part 2-a review of ontology mapping and evolving. 16 & Services. ACM, 317–324. 16 Journal of information science 28, 5 (2002), 375–388. [34] Luke Georghiou. 1998. Global cooperation in research. Re- 17 17 [18] AnHai Doan, Jayant Madhavan, Pedro Domingos, and Alon search policy 27, 6 (1998), 611–626. 18 Halevy. 2002. Learning to map between ontologies on the se- [35] Wolfgang Glänzel. 2000. Science in Scandinavia: A bibliomet- 18 19 mantic web. In Proceedings of the 11th international confer- ric approach. Scientometrics 48, 2 (2000), 121–150. 19 ence on World Wide Web. AcM, 662–673. 20 [36] Roger Guimera, Brian Uzzi, Jarrett Spiro, and Luis A Nunes 20 [19] AnHai Doan, Jayant Madhavan, Pedro Domingos, and Alon 21 Amaral. 2005. Team assembly mechanisms determine collab- 21 Halevy. 2004. Ontology matching: A machine learning ap- oration network structure and team performance. Science 308, 22 proach. In Handbook on ontologies. Springer, 385–403. 22 5722 (2005), 697–702. 23 [20] Zlatan Dragisic, Valentina Ivanova, Huanyu Li, and Patrick 23 [37] Ian Harrow, Ernesto Jiménez-Ruiz, Andrea Splendiani, Mar- 24 Lambrix. 2017. Experiences from the anatomy track in the on- 24 tin Romacker, Peter Woollard, Scott Markel, Yasmin Alam- tology alignment evaluation initiative. Journal of biomedical 25 Faruque, Martin Koch, James Malone, and Arild Waaler. 2017. 25 semantics 8, 1 (2017), 56. 26 Matching disease and phenotype ontologies in the ontology 26 [21] Amir Ebrahimi Fard and Cunningham Scott. [n. d.]. Assessing alignment evaluation initiative. Journal of biomedical seman- 27 the Readiness of the Academia in the Topic of False and Un- 27 tics 8, 1 (2017), 55. 28 verified Information. To be appeared in ACM Journal of Data 28 [38] Yulan He and Siu Cheung Hui. 2002. Mining a web citation 29 and Information Quality (JDIQ) ([n. d.]). 29 database for author co-citation analysis. Information process- 30 [22] Marc Ehrig. 2006. Ontology alignment: bridging the semantic 30 ing & management 38, 4 (2002), 491–508. gap. Vol. 4. Springer Science & Business Media. 31 [39] David Hsiehchen, Magdalena Espinoza, and Antony Hsieh. 31 [23] Islam Elgedawy. 2003. A conceptual framework for web ser- 32 32 vices semantic discovery. In OTM Confederated International 2015. Multinational teams and diseconomies of scale in col- 33 Conferences" On the Move to Meaningful Internet Systems". laborative research. Science advances 1, 8 (2015), e1500211. 33 34 Springer, 1004–1016. [40] Jingshan Huang, Jiangbo Dang, José M Vidal, and Michael N 34 35 [24] Scott Emmons, Stephen Kobourov, Mike Gallant, and Katy Huhns. 2007. Ontology matching using an artificial neural net- 35 work to learn weights. In IJCAI workshop on semantic Web for 36 Börner. 2016. Analysis of network clustering algorithms 36 and cluster quality metrics at scale. PloS one 11, 7 (2016), collaborative knowledge acquisition, Vol. 106. 37 37 e0159161. [41] Bernhard G Humm and Janina Fengel. 2012. Semantics-based 38 38 [25] Vadim Ermolayev and Maxim Davidovsky. 2012. Agent-based business process model similarity. In International Conference 39 ontology alignment: basics, applications, theoretical founda- on Business Information Systems. Springer, 36–47. 39 40 tions, and demonstration. In Proceedings of the 2nd Interna- [42] Prateek Jain, Pascal Hitzler, Amit P Sheth, Kunal Verma, and 40 41 tional Conference on Web Intelligence, Mining and Semantics. Peter Z Yeh. 2010. Ontology alignment for linked open data. 41 In International semantic web conference. Springer, 402–417. 42 ACM, 3. 42 [26] Jérôme Euzenat, Pavel Shvaiko, et al. 2007. Ontology match- [43] Ernesto Jiménez-Ruiz, Asan Agibetov, Matthias Samwald, and 43 43 ing. Vol. 18. Springer. Valerie Cross. 2018. Breaking-down the Ontology Alignment 44 [27] Daniel Faria, Catia Pesquita, Emanuel Santos, Matteo Pal- Task with a Lexical Index and Neural Embeddings. arXiv 44 45 monari, Isabel F Cruz, and Francisco M Couto. 2013. The preprint arXiv:1805.12402 (2018). 45 46 agreementmakerlight ontology matching system. In OTM Con- [44] Ernesto Jiménez-Ruiz, Asan Agibetov, Matthias Samwald, and 46 47 federated International Conferences" On the Move to Mean- Valerie Cross. 2018. Breaking-down the Ontology Alignment 47 ingful Internet Systems" Task with a Lexical Index and Neural Embeddings. arXiv 48 . Springer, 527–541. 48 [28] Aissa Fellah, Mimoun Malki, and Atilla Elçi. 2016. Web preprint arXiv:1805.12402 (2018). 49 49 services matchmaking based on a partial ontology alignment. [45] Ernesto Jiménez-Ruiz and Bernardo Cuenca Grau. 2011. 50 Int. Journal of Information Technology and Computer Science Logmap: Logic-based and scalable ontology matching. In In- 50 51 (IJITCS) 8, 6 (2016). ternational Semantic Web Conference. Springer, 273–288. 51 28 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative

1 [46] Ernesto Jiménez-Ruiz, Christian Meilicke, Bernardo Cuenca [63] David Martin, Mark Burstein, Jerry Hobbs, Ora Lassila, 1 2 Grau, and Ian Horrocks. 2013. Evaluating Mapping Repair Drew McDermott, Sheila McIlraith, Srini Narayanan, Mas- 2 3 Systems with Large Biomedical Ontologies. Description Log- simo Paolucci, Bijan Parsia, Terry Payne, et al. 2004. OWL-S: 3 ics 13 (2013), 246–257. Semantic markup for web services. W3C member submission 4 4 [47] Jason J Jung. 2009. Semantic business process integration 22, 4 (2004). 5 based on ontology alignment. Expert Systems with Applica- [64] Christian Meilicke. 2011. Alignment incoherence in ontology 5 6 tions 36, 8 (2009), 11013–11020. matching. Ph.D. Dissertation. Universität Mannheim. 6 7 [48] Yannis Kalfoglou and Marco Schorlemmer. 2003. IF-Map: An [65] Christian Meilicke, Heiner Stuckenschmidt, and Andrei 7 8 ontology-mapping method based on information-flow theory. Tamilin. 2007. Repairing ontology mappings. In AAAI, Vol. 3. 8 9 In Journal on data semantics I. Springer, 98–127. 6. 9 [49] Yannis Kalfoglou and Marco Schorlemmer. 2003. IF-Map: An [66] Prasenjit Mitra, Natasha F Noy, and Anuj Rattan Jaiswal. 2005. 10 10 ontology-mapping method based on information-flow theory. Omen: A probabilistic ontology mapping tool. In International 11 In Journal on data semantics I. Springer, 98–127. Semantic Web Conference. Springer, 537–547. 11 12 [50] Yannis Kalfoglou and Marco Schorlemmer. 2003. IF-Map: An [67] Majid Mohammadi. 2019. Bayesian Evaluation and Compari- 12 13 ontology-mapping method based on information-flow theory. son of Ontology Alignment Systems. IEEE Access (2019). 13 14 In Journal on data semantics I. Springer, 98–127. [68] Majid Mohammadi, Amir Ahooye Atashin, Wout Hofman, and 14 [51] Yannis Kalfoglou and Marco Schorlemmer. 2003. Ontology Yaohua Tan. 2018. Comparison of ontology alignment systems 15 15 mapping: the state of the art. The knowledge engineering re- across single matching task via the McNemarâA˘ Zs´ test. ACM 16 view 18, 1 (2003), 1–31. Transactions on Knowledge Discovery from Data (TKDD) 12, 16 17 [52] Qing Ke, Emilio Ferrara, Filippo Radicchi, and Alessandro 4 (2018), 51. 17 18 Flammini. 2015. Defining and identifying sleeping beauties in [69] Majid Mohammadi, Wout Hofman, and Yao-Hua Tan. 2019. 18 19 science. Proceedings of the National Academy of Sciences 112, A comparative study of ontology matching systems via infer- 19 20 24 (2015), 7426–7431. ential statistics. IEEE Transactions on Knowledge and Data 20 [53] Hak Lae Kim, Alexandre Passant, John G Breslin, Simon Engineering 31, 4 (2019), 615–628. 21 21 Scerri, and Stefan Decker. 2008. Review and alignment of [70] Majid Mohammadi, Wout Hofman, and Yao-Hua Tan. 2019. 22 tag ontologies for semantically-linked data in collaborative tag- Simulated annealing-based ontology matching. ACM Trans- 22 23 ging spaces. In 2008 IEEE International Conference on Se- actions on Management Information Systems (TMIS) 10, 1 23 24 mantic Computing. IEEE, 315–322. (2019), 3. 24 25 [54] Christopher Klinkmüller, Ingo Weber, Jan Mendling, Henrik [71] Mathias Niepert, Christian Meilicke, and Heiner Stucken- 25 Leopold, and André Ludwig. 2013. Increasing recall of pro- schmidt. 2010. A probabilistic-logical framework for ontol- 26 26 cess model matching by improved activity label matching. In ogy matching. In Twenty-Fourth AAAI Conference on Artificial 27 Business process management. Springer, 211–218. Intelligence. 27 28 [55] Manel Kolli and Zizette Boufaida. 2011. A description logics [72] Henrik Nottelmann and Umberto Straccia. 2006. A probabilis- 28 29 formalization for the ontology matching. Procedia Computer tic, logic-based framework for automated web directory align- 29 30 Science 3 (2011), 29–35. ment. In Soft Computing in Ontologies and Semantic Web. 30 31 [56] Aliya Kuzhabekova and Jennifer Kuzma. 2014. Mapping the Springer, 47–77. 31 emerging field of genome editing. Technology Analysis & [73] Natalya Fridman Noy and Mark A Musen. 1999. SMART: Au- 32 32 Strategic Management 26, 3 (2014), 321–352. tomated support for ontology merging and alignment. In Proc. 33 [57] Loredana Laera, Valentina Tamma, Jérôme Euzenat, Trevor of the 12th Workshop on Knowledge Acquisition, Modelling, 33 34 Bench-Capon, and Terry Payne. 2006. Reaching agreement and Management (KAW’99), Banf, Canada. Citeseer. 34 35 over ontology alignments. In International Semantic Web Con- [74] Natalya F Noy and Mark A Musen. 2003. The PROMPT suite: 35 36 ference. Springer, 371–384. interactive tools for ontology merging and mapping. Interna- 36 [58] Nadira Lammari, Jacky Akoka, and Isabelle Comyn-Wattiau. tional journal of human-computer studies 59, 6 (2003), 983– 37 37 2003. Supporting database evolution: Using ontologies match- 1024. 38 38 ing. In International Conference on Object-Oriented Informa- [75] Natalya Fridman Noy, Mark A Musen, et al. 2000. Algorithm 39 tion Systems. Springer, 284–288. and tool for automated ontology merging and alignment. In 39 40 [59] Loet Leydesdorff and Ping Zhou. 2007. Nanotechnology as a Proceedings of the 17th National Conference on Artificial In- 40 41 field of science: Its delineation in terms of journals and patents. telligence (AAAI-00). Available as SMI technical report SMI- 41 42 Scientometrics 70, 3 (2007), 693–713. 2000-0831, Vol. 115. sn. 42 [60] Lu Liu, Yang Wang, Roberta Sinatra, C Lee Giles, Chaoming [76] Leo Obrst, Robert E Wray, and Howard Liu. 2001. Ontolog- 43 43 Song, and Dashun Wang. 2018. Hot streaks in artistic, cultural, ical engineering for B2B E-commerce. In Proceedings of the 44 and scientific careers. Nature 559, 7714 (2018), 396. international conference on Formal Ontology in Information 44 45 [61] Ajoy Mallik and Nripendranath Mandal. 2014. Bibliometric Systems-Volume 2001. ACM, 117–126. 45 46 analysis of global publication output and collaboration struc- [77] Lorena Otero-Cerdeira, Francisco J Rodríguez-Martínez, and 46 47 ture study in microRNA research. Scientometrics 98, 3 (2014), Alma Gómez-Rodríguez. 2015. Ontology matching: A liter- 47 Expert Systems with Applications 48 2011–2037. ature review. 42, 2 (2015), 48 [62] Ming Mao, Yefei Peng, and Michael Spring. 2010. An adap- 949–971. 49 49 tive ontology mapping approach with neural network based [78] Catia Pesquita, Daniel Faria, Emanuel Santos, and Francisco M 50 constraint satisfaction. Web Semantics: Science, Services and Couto. 2013. To repair or not to repair: reconciling correctness 50 51 Agents on the World Wide Web 8, 1 (2010), 14–25. and coherence in ontology reference alignments. In Proc. 8th 51 M. Mohammadi and A. Ebrahimi Fard / Ontology Alignment Revisited: A Bibliometric Narrative 29

1 ISWC ontology matching workshop (OM), Sydney (AU), page [91] Evren Sirin, Bijan Parsia, and James Hendler. 2004. Filtering 1 2 this volume. and selecting semantic web services with interactive composi- 2 3 [79] Alexander M Petersen, Dinesh Majeti, Kyeongan Kwon, Mo- tion techniques. IEEE Intelligent Systems 19, 4 (2004), 42–49. 3 hammed E Ahmed, and Ioannis Pavlidis. 2018. Cross- 4 [92] Jie Tang, Juanzi Li, Bangyong Liang, Xiaotong Huang, Yi Li, 4 disciplinary evolution of the genomics revolution. Science ad- and Kehong Wang. 2006. Using Bayesian decision for ontol- 5 vances 4, 8 (2018), eaat4211. 5 ogy mapping. Web Semantics: Science, Services and Agents on 6 [80] Guy Pierra. 2004. The PLIB ontology-based approach to data 6 the World Wide Web 4, 4 (2006), 243–262. 7 integration. In Building the Information Society. Springer, 13– 7 [93] Mike Thelwall, Stefanie Haustein, Vincent Larivière, and Cas- 8 18. 8 [81] Alan L Porter, Jan Youtie, Philip Shapira, and David J Schoe- sidy R Sugimoto. 2013. Do altmetrics work? Twitter and ten 9 9 neck. 2008. Refining search terms for nanotechnology. Journal other social web services. PloS one 8, 5 (2013), e64841. 10 of nanoparticle research 10, 5 (2008), 715–728. [94] Elodie Thiéblin, Ollivier Haemmerlé, Nathalie Hernandez, and 10 11 [82] Martin F Porter. 1980. An algorithm for suffix stripping. Pro- Cassia Trojahn. 2019. Survey on complex ontology matching. 11 12 gram 14, 3 (1980), 130–137. Semantic Web Preprint (2019), 1–39. 12 13 [83] George Radda. 2002. Biomedical research and international [95] Nees van Eck and Ludo Waltman. 2009. Software survey: 13 collaboration. Science 295, 5554 (2002), 445–446. 14 VOSviewer, a computer program for bibliometric mapping. 14 [84] Mariano Rubiolo, María Laura Caliusco, Georgina Stegmayer, Scientometrics 84, 2 (2009), 523–538. 15 Mauricio Coronel, and M Gareli Fabrizi. 2012. Knowledge 15 [96] A Van Raan. 1998. The influence of international collaboration 16 discovery through ontology matching: An approach based on 16 on the impact of research results: Some simple mathematical 17 an Artificial Neural Network model. Information Sciences 194 17 considerations concerning the role of self-citations. Sciento- 18 (2012), 107–119. 18 [85] Ulrich Schmoch and Torben Schubert. 2008. Are international metrics 42, 3 (1998), 423–428. 19 19 co-publications an indicator for quality of scientific research? [97] Matthias Weidlich, Tomer Sagi, Henrik Leopold, Avigdor Gal, 20 Scientometrics 74, 3 (2008), 361–377. and Jan Mendling. 2013. Predicting the quality of process 20 21 [86] Marco Schorlemmer, Yannis Kalfoglou, and Manuel Atencia. model matching. In Business Process Management. Springer, 21 22 2007. A formal foundation for ontology-alignment interaction 203–210. 22 23 models. International Journal on Semantic Web and Informa- [98] Floris Wiesman, Nico Roos, and Paul Vogt. 2002. Automatic 23 tion Systems (IJSWIS) 3, 2 (2007), 50–68. ontology mapping for agent communication. In Proceedings of 24 24 [87] Aviv Segev and Eran Toch. 2009. Context-based matching and the first international joint conference on Autonomous agents 25 25 ranking of web services for composition. IEEE Transactions and multiagent systems: part 2. ACM, 563–564. 26 on Services Computing 26 3 (2009), 210–222. [99] Claes Wohlin, Aybuke Aurum, Lefteris Angelis, Laura Phillips, [88] Per O Seglen and Dag W Aksnes. 2000. Scientific productivity 27 Yvonne Dittrich, Tony Gorschek, Hakan Grahn, Kennet Hen- 27 and group size: A bibliometric analysis of Norwegian microbi- 28 ningsson, Simon Kagstrom, Graham Low, et al. 2011. The suc- 28 ological research. Scientometrics 49, 1 (2000), 125–143. 29 cess factors powering industry-academia collaboration. IEEE 29 [89] Pavel Shvaiko and Jérôme Euzenat. 2013. Ontology matching: 30 state of the art and future challenges. IEEE Transactions on software 29, 2 (2011), 67–73. 30 31 knowledge and data engineering 25, 1 (2013), 158–176. [100] Michel Zitt and Elise Bassecoulard. 2006. Delineating com- 31 32 [90] Nuno Silva, João Rocha, and José Cardoso. 2003. E-Business plex scientific fields by an hybrid lexical-citation method: An 32 Information processing & man- 33 interoperability through ontology semantic mapping. In Work- application to nanosciences. 33 ing Conference on Virtual Enterprises. Springer, 315–322. agement 42, 6 (2006), 1513–1531. 34 34 35 35 36 36 37 37 38 38 39 39 40 40 41 41 42 42 43 43 44 44 45 45 46 46 47 47 48 48 49 49 50 50 51 51