DOCSLIB.ORG

Large Scale Record Linkage in the Presence of Missing Data Otherwise (A8,A 9 ) Is a Member of U (True Non-Matches) and A8 and Definition 3.3 (Relational Signature)

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Large Scale Record Linkage in the Presence of Missing Data Otherwise (A8,A 9 ) Is a Member of U (True Non-Matches) and A8 and Definition 3.3 (Relational Signature) Large Scale Record Linkage in the Presence of Missing Data Thilina Ranbaduge Peter Christen Rainer Schnell The Australian National University The Australian National University University Duisburg-Essen Canberra, Australia Canberra, Australia Duisburg, Germany [email protected] [email protected] [email protected] ABSTRACT One major data quality aspect that so far has only seen limited Record linkage is aimed at the accurate and efficient identifica- attention in RL is missing data [4, 16, 19, 28]. There are various tion of records that represent the same entity within or across reasons why missing values can occur, ranging from equipment disparate databases. It is a fundamental task in data integration malfunction or data items not considered to be important, to dele- and increasingly required for accurate decision making in appli- tion of values due to inconsistencies, or even the refusal of individ- cation domains ranging from health analytics to national security. uals to provide information for example when answering surveys. Traditional record linkage techniques calculate string similarities Missing data can be categorised into different types [26]. between quasi-identifying (QID) values, such as the names and ad- Data missing completely at random (MCAR) are missing values dresses of people. Errors, variations, and missing QID values can that occur without any patterns or correlations at all with any however lead to low linkage quality because the similarities be- other values in the same record or database. For example, if a first tween records cannot be calculated accurately. To overcome this name is missing for an individual, then neither last name, address, challenge, we propose a novel technique that can accurately link age, nor gender, can help predict the missing first name value (as- records even when QID values contain errors or variations, or are suming no external database is accessible where a complete record missing. We first generate attribute signatures (concatenated QID of that person is available). With data missing at random (MAR) one values) using an Apriori based selection of suitable QID attributes, can assume that a missing value can be predicted by other values in and then relational signatures that encapsulate relationship infor- the same record and/or database. As an example, for a record of a mation between records. Combined, these signatures can uniquely surgeon in an employment database, her salary could be predicted identify individual records and facilitate fast and high quality link- by averaging the salaries of all other (female) surgeons in that data- ing of very large databases through accurate similarity calculations base. Data missing not at random (MNAR) do occur for some spe- between records. We evaluate the linkage quality and scalability of cific reasons, for example if a patient in a medical study suffered our approach using large real-world databases, showing that it can from a stroke and therefore did not return to re-examinations then achieve high linkage quality even when the databases being linked there will be missing values for this patient in the study’s data- contain substantial amounts of missing values and errors. base. Finally, structurally missing data are values that are missing because they should not exist, and missing is their correct value. For example, young children should not have an occupation. Struc- 1 INTRODUCTION turally missing data are different to MNAR data because for the Organisations such as financial institutions, statistical agencies, and former the correct actual value is an empty value, while for MNAR government departments, increasingly require records about en- data there does exist a correct value, but not recorded. tities from multiple databases to be integrated to allow efficient Missing data can often occur in the QID attributes used to link and accurate decision making [12], where such databases can con- databases. While data imputation can be applied with the aim to tain many millions of records with detailed information about peo- fill such missing values before the databases are linked [22, 26], in ple, such as customers or patients [6, 18]. Record linkage (RL) [15] our work we assume that not all missing values have been imputed. is one major task required when databases are to be integrated. Specifically, imputation is not possible for missing values of types arXiv:2104.09677v1 [cs.DB] 19 Apr 2021 RL aims to identify and match records that refer to the same enti- MCAR and structurally missing. For example, as shown in Table 1, ties across different databases [6]. Linked databases allow improve- if the street address is missing for an individual (like forA1), then no ment of data quality, enrichment of the information known about other QID value of this record can help to predict this missing last entities, and facilitate the discovery of novel patterns and relation- name value. Missing values can lead to lower similarities between ships between entities that cannot be identified from individual records and thereby affect linkage quality. If alternatively those databases [7, 12, 32]. records or QIDs with missing values are not used for a linkage at Because there is often a lack of unique entity identifiers (such all, then linkage quality will also likely suffer. as social security numbers) across the databases to be linked, RL As an example, given the five records of three entities in the is commonly based on partially identifying attributes, known as two databases D and D in Table 1, a linkage between D and quasi-identifiers (QIDs) [8], such as names, addresses, dates of birth, D would potentially classify the record pairs (A1, A4), and (A2, A4) and so on. Data quality aspects such as typographical errors, vari- as matches because they have similar QID values, and it would ations, and changes of values over time (for example when people potentially classify records A3 and A5 to refer to different entities move or change their names due to marriage) are common in many because they have a low similarity due to name variations and of the QID attributes used for RL [6]. Due to errors and variations missing values. Therefore, this linkage will be of poor quality due in QIDs, exact matching of attribute values can lead to poor linkage to the wrongly matched record pair (A1, A4) and the non-matched quality [6, 8]. true matching record pair (A3, A5). Thilina Ranbaduge, Peter Christen, and Rainer Schnell Table 1: Two small example databases, D and D , with missing values and variations, and where RecordID and EntityID represent record and entity identifiers, respectively. Database RecordID EntityID FirstName LastName BirthDate StreetAddress City PhoneNumber A1 41 Peter Smith 1981-11-25 London D A2 42 Peter 1981-11-25 43SkyePl Dublin +353456785 A3 43 Anne Miller 1991-09-11 43SkyePlc +353456785 A4 42 Peter Smith 1981-11-25 43SkyePl D A5 43 Ann Myller 43SkyePlace Dublin +353456785 Contribution: We propose a novel RL technique that is appli- the EM algorithm, but does not scale to large databases due to its cable in situations where QID values contain errors and variations, computationally expensive weight estimations. and importantly that can be missing. For each record in a database Goldstein and Harron [19] developed an approach to link at- to be linked, our approach first generates several distinct attribute tributes of interest for a statistical model between a primary data- signatures by concatenating values from a set of QID attributes [34]. base and one or more secondary databases that contain the at- We propose an Apriori [5] based technique to select suitable QID tributes of interest. The approach exploits relationships between attributes for signature generation. As we detail in Section 4, the attributes that are shared across the databases, and treats the link- aim of attribute signatures is to uniquely describe records such that age of these databases as a missing data problem. The approach record pairs can be matched based on the number of common at- uses imputation to combine information from records in the dif- tribute signatures they share. ferent databases that belong to the same individual, and calculates We then utilise the relationships between entities to identify match weights to correct selection bias. matching records assuming they contain missing or dirty values in Ferguson et al. [16] proposed an approach to improve RL when their QID attributes. We first generate a graph of records based on the databases to be linked contain missing values by using a modi- the relationships between them, for example the roles of individu- fication ofthe EM algorithm [9] that considers boththe imputation als in a census household (such as father of or spouse of ), or shared of missing values as well as correlations between QID attribute val- attribute values such as addresses or phone numbers. We then gen- ues. Aninya et al. [4] analysed how different blocking techniques erate a relational signature for each record based on its neighbour- are affected by missing values. They showed that techniques that hood in this graph, and calculate the similarities between relational insert each record into multiple blocks, such as canopy clustering signatures of pairs of records. and suffix array indexing [6], performed best when the attributes We analyse our approach in terms of its computational complex- used for blocking contained missing values. ity, and evaluate linkage quality using large real-world data sets, In contrast to these existing methods [4, 16, 19, 28], our ap- including simulated German census databases containing in total proach does not consider imputation or estimation of similarities over 130 million records. As the experimental results in Section 5 when QID values are missing. Instead, we exploit relational in- show, our approach scales to large databases and incorporating re- formation between records to increase the overall similarities be- lational signatures can help to substantially improve linkage qual- tween record pairs even if their attribute similarities are low due ity compared to several baseline approaches [13, 15, 28, 34].
Load more

Recommended publications
  • A Framework for Detecting Unnecessary Industrial Data in ETL Processes
    View metadata, citation and similar papers at core.ac.uk brought to you by CORE provided by Apollo A Framework for Detecting Unnecessary Industrial Data in ETL Processes Philip Woodall Duncan McFarlane Department of Engineering Department of Engineering University of Cambridge, UK. University of Cambridge, UK. [email protected] Amar Shah, Department of Engineering Torben Jess University of Cambridge, UK Department of Engineering University of Cambridge, UK. William Krechel Boeing, USA Mark Harrison Eric Nicks Department of Engineering Boeing, USA. University of Cambridge, UK Abstract—Extract, transform and load (ETL) is a critical procurement system that is used to select suppliers for the process used by industrial organisations to shift data from one orders of the parts. These orders may then also be transferred database to another, such as from an operational system to a data by ETL into the order release system, ready to send to the warehouse. With the increasing amount of data stored by suppliers. Moreover, for reporting, these orders may be sent via industrial organisations, some ETL processes can take in excess ETL to a data warehouse that is used by managers to, for of 12 hours to complete; this can leave decision makers stranded example, scrutinise various orders over the last month. while they wait for the data needed to support their decisions. After designing the ETL processes, inevitably data requirements However, the problem is that ETL is a time-consuming and can change, and much of the data that goes through the ETL expensive process, which can leave decision makers in process may not ever be used or needed.
    [Show full text]
  • Historical Census Record Linkage
    Historical Census Record Linkage Steven Ruggles† University of Minnesota Catherine Fitch University of Minnesota Evan Roberts University of Minnesota December 2017 Working Paper No. 2017-3 https://doi.org/10.18128/ MPC2017-3 †Correspondence should be directed to: Steven Ruggles University of Minnesota, 50 Willey Hall, 225 19th Ave S., Minneapolis, MN 55455 e-mail: [email protected], phone: 612-624-5818, fax:612-626-8375 Abstract For the past 80 years, social scientists have been linking historical censuses across time to study economic and geographic mobility. In recent decades, the quantity of historical census record linkage has exploded, owing largely to the advent of new machine-readable data created by genealogical organizations. Investigators are examining economic and geographic mobility across multiple generations, but also engaging many new topics. Several analysts are exploring the effects of early-life socioeconomic conditions, environmental exposures, or natural disasters on family, health and economic outcomes in later life. Other studies exploit natural experiments to gauge the impact of policy interventions such as social welfare programs and educational reforms. The new data sources have led to a proliferation of record linkage methodology, and some widespread approaches inadvertently introduce errors that can lead to false inferences. A new generation of large-scale shared data infrastructure now in preparation will ameliorate weaknesses of current linkage methods. Introduction Historical census data are indispensable for studying long-run change, since they provide the only record of the lives of millions of people over the past two centuries. In much of Europe and North America, investigators can access individual-level data based on original census manuscripts (Szołtysek & Gruber 2016).
    [Show full text]
  • Volume I, Record Linkage Issues and Resul
    Chapter Two Administrative Data and Record Linkage Issues For research purposes, administrative data have the advantage of detailed and accurate measurement of program status and outcomes, complete coverage of populations of interest (enabling detailed subgroup analyses), data on the same individuals over long periods, low cost relative to survey data, and the ability to obtain many kinds of information through matching (Hotz et al., 1998). In addition, many types of administrative data have relatively high degrees of uniformity in content across geographic areas.13,14 Government agencies have recognized the potential research uses of administrative data. Of 10 data development initiatives recently identified by USDA’s Economic Research Service, only one did not involve administrative data (Wittenburg et al., 2001). Five of the 10 initiatives involve creation of linked databases matching administrative records from multiple agencies, or matching administrative records to survey responses. Linked databases are a way to create “new” data from existing sources. For example, the National Center for Health Statistics (NCHS) determines infant mortality rates using state files of linked data from birth and death certificates (Mathews et al., 2002). The Department of Transportation examines motor vehicle crash outcomes by linking records of police-reported crashes to hospital discharge data, EMS data, and hospital emergency department data (NHTSA, 1996a). In the social services arena, a number of States have developed master client indexes that
    [Show full text]
  • Approaches to Multiple Record Linkage
    Approaches to Multiple Record Linkage Mauricio Sadinle, Rob Hall, and Stephen E. Fienberg Department of Statistics, Heinz College, and Department of Machine Learning Carnegie Mellon University Pittsburgh, PA 15213-3890, U.S.A. E-mail: [email protected]; [email protected]; fi[email protected] We review the theory and techniques of record linkage that date back to pioneering work by Fellegi and Sunter on matching records( in) two lists. When the task involves linking K > 2 lists, the most common approach K consists of performing all 2 possible pairs of lists using a Fellegi-Sunter-like approach and then somehow reconciling the discrepancies in an ad hoc fashion. We describe some important uses of the methodology, provide a principled way of accomplishing the reconciliation and we finally present some key parts of the generalization of Fellegi and Sunter's method to K > 2 lists. Introduction Record linkage is a family of techniques for matching two data files using names, addresses, and other fields that are typically not unique identifiers of entities. Most record linkage approaches are based or emulate a method presented in a pioneering paper by Fellegi and Sunter (1969). Winkler (1999) and Herzog et al. (2007) document the basic two-list methodology and several variations. We begin by reviewing some common applications of record linkage techniques, including the linking K > 2 lists. Data Integration. Synthesizing data on a group of individuals from two or more files in the absence of unique identifiers requires record linkage, e.g., to create a data warehouse to be used for querying such as credit checks, e.g., see Talburt (2011), or to assess enrollment in multiple government programs, e.g., see Cole (2003).
    [Show full text]
  • Record Linkage: a Machine Learning Approach, a Toolbox, and a Digital Government Web Service
    Purdue University Purdue e-Pubs Department of Computer Science Technical Reports Department of Computer Science 2003 Record Linkage: A Machine Learning Approach, A Toolbox, and a Digital Government Web Service Mohamed G. Elfeky Vassilios S. Verykios Ahmed K. Elmagarmid Purdue University, [email protected] Thanaa M. Ghanem Ahmed R. Huwait Report Number: 03-024 Elfeky, Mohamed G.; Verykios, Vassilios S.; Elmagarmid, Ahmed K.; Ghanem, Thanaa M.; and Huwait, Ahmed R., "Record Linkage: A Machine Learning Approach, A Toolbox, and a Digital Government Web Service" (2003). Department of Computer Science Technical Reports. Paper 1573. https://docs.lib.purdue.edu/cstech/1573 This document has been made available through Purdue e-Pubs, a service of the Purdue University Libraries. Please contact [email protected] for additional information. RECORD LINKAGE, A MACHINE LEARNING APPROACH, A TOOLBOX, AND A DIGITAL GOVERNMENT WEB SERVICE Mohamed G. Elfeky Vassilios S. Verykios Ahmed K. Elmagarmid Thanaa M. Ghanem Ahmed R. Huwait CSD TR #03-024 July 2003 Record Linkage: A Machine Learning Approach, A Toolbox, and A Digital Government Web Service' l Mohamed G. Elfeky 1 VassiliosS. Verykios Ahmed K. Elmagarrnid 1 Purdue University Drexel University Hewlett-Packard [email protected] [email protected] [email protected] Thanaa M. Ghanem Ahmed R. Huwait 2 Purdue University Ain Shams University [email protected],edu [email protected] Abstract Data cleaning is a vital process mat ensures the quality of data smred in real·world databases. Data cleaning problems are frequently encountered in many research areas, such as kllowledge discovery in databases, data warellOusing. system imegratioll and e­ services.
    [Show full text]
  • Contributions to Record Linkage for Disclosure Risk Assessment
    Contributions to Record Linkage for Disclosure Risk Assessment by Jordi Nin Guerrero Advisor: Dr. Vicenç Torra i Reventós A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor in Artificial Intelligence at the Departament de Ciències de la Computació Universitat Autònoma de Barcelona A la Mireia , que fa seves les meves il·lusions. I a en Xavier, a qui esperem amb entusiasme. ii I am only a child playing on the beach, while vast oceans of truth lie undiscovered before me. Isaac Newton (1642-1727) iii Contents Agraïments xiii Abstract xv 1 Introduction 1 1.1 Motivations ................................... 1 1.2 Contributions .................................. 3 1.3 StructureoftheDocument . 4 2 Preliminaries 7 2.1 AggregationFunctions . 7 2.2 TimeSeries.................................... 12 2.3 Re-identification Methods . 17 2.4 Microdata Protection Methods . 21 2.5 Information Loss and Disclosure Risk . 33 2.6 DataSetsDescription ............................. 38 3 Microaggregation Analysis 45 3.1 Attribute Selection in Multivariate Microaggregation ........... 45 3.2 Modeling Projections in Microaggregation . ...... 58 3.3 Improving Microaggregation for Complex Record Anonymization . 62 v 4 Specific Disclosure Risk Measures 77 4.1 Rank Swapping Record Linkage . 77 4.2 AlignmentRecordLinkage. 93 4.3 ProjectedRecordLinkage . 98 5 Record Linkage using Fuzzy Integrals 109 5.1 An Alternative Disclosure Risk Scenario . 109 5.2 Experiments ................................... 117 6 Time Series Protection 125 6.1 TimeSeriesProtection .. .. .. .. .. .. 125 6.2 Time Series Information Loss Measures . 127 6.3 Time Series Disclosure Risk Measures . 130 6.4 Final Trade-off Evaluation . 135 6.5 Experiments ................................... 136 7 Conclusions and Future Directions 141 7.1 Summary of Contributions .
    [Show full text]
  • Neural Networks for Entity Matching: a Survey
    Neural Networks for Entity Matching: A Survey NILS BARLAUG, Cognite, Norway and NTNU, Norway JON ATLE GULLA, NTNU, Norway Entity matching is the problem of identifying which records refer to the same real-world entity. It has been actively researched for decades, and a variety of different approaches have been developed. Even today, it remains a challenging problem, and there is still generous room for improvement. In recent years we have seen new methods based upon deep learning techniques for natural language processing emerge. In this survey, we present how neural networks have been used for entity matching. Specifically, we identify which steps of the entity matching process existing work have targeted using neural networks, and provide an overview of the different techniques used at each step. We also discuss contributions from deep learning in entity matching compared to traditional methods, and propose a taxonomy of deep neural networks for entity matching. CCS Concepts: • Computing methodologies → Neural networks; Natural language processing; • Infor- mation systems → Entity resolution. Additional Key Words and Phrases: deep learning, entity matching, entity resolution, record linkage, data matching ACM Reference Format: Nils Barlaug and Jon Atle Gulla. 2021. Neural Networks for Entity Matching: A Survey. ACM Trans. Knowl. Discov. Data. 15, 3 (April 2021), 36 pages. https://doi.org/10.1145/3442200 1 INTRODUCTION Our world is becoming increasingly digitalized. While this opens up a number of new, exciting opportunities, it also introduces challenges along the way. A substantial amount of the value to be harvested from increased digitalization depends on integrating different data sources. Unfortunately, many of the existing data sources one wishes to integrate do not share a common frame of reference.
    [Show full text]
  • Introduction to Data Science Rest of Today’S Lecture
    INTRODUCTION TO DATA SCIENCE REST OF TODAY’S LECTURE Exploratory Analysis, Insight & Data Data hypothesis analysis Policy collection processing & testing, & Data viz ML Decision Continue with the general topic of data wrangling and cleaning Many slides from Amol Deshpande (UMD) ! 2 OVERVIEW Goal: get data into a structured form suitable for analysis • Variously called: data preparation, data munging, data curation • Also often called ETL (Extract-Transform-Load) process Often the step where majority of time (80-90%) is spent Key steps: • Scraping: extracting information from sources, e.g., webpages, spreadsheets • Data transformation: to get it into the right structure • Data integration: combine information from multiple sources • Information extraction: extracting structured information from unstructured/text sources • Data cleaning: remove inconsistencies/errors ! 3 OVERVIEW Goal: get data into a structured form suitable for analysis • Variously called: data preparation, data munging, data curation • Also often called ETL (Extract-Transform-Load) process Often the step where majority of time (80-90%) is spent Key steps: Already • Scraping: extracting information from sources, e.g., webpages,covered spreadsheets • Data transformation: to get it into the right structure • Information extraction: extracting structured information from unstructured/text sources In a few • Data integration: combine information from multiple sourcesclasses • Data cleaning: remove inconsistencies/errors ! 4 OVERVIEW !5 OVERVIEW Many of the problems are not
    [Show full text]
  • Learning Blocking Schemes for Record Linkage∗
    Learning Blocking Schemes for Record Linkage∗ Matthew Michelson and Craig A. Knoblock University of Southern California Information Sciences Institute, 4676 Admiralty Way Marina del Rey, CA 90292 USA {michelso,knoblock}@isi.edu Abstract Record linkage is the process of matching records across data sets that refer to the same entity. One issue within record linkage is determining which record pairs to consider, since a detailed comparison between all of the records is imprac- tical. Blocking addresses this issue by generating candidate matches as a preprocessing step for record linkage. For ex- ample, in a person matching problem, blocking might return all people with the same last name as candidate matches. Two main problems in blocking are the selection of attributes for generating the candidate matches and deciding which meth- ods to use to compare the selected attributes. These attribute and method choices constitute a blocking scheme. Previ- ous approaches to record linkage address the blocking issue in a largely ad-hoc fashion. This paper presents a machine learning approach to automatically learn effective blocking Figure 1: Record linkage example schemes. We validate our approach with experiments that show our learned blocking schemes outperform the ad-hoc blocking schemes of non-experts and perform comparably to or cluster the data sets by the attribute. Then we apply the those manually built by a domain expert. comparison method to only a single member of a block. Af- ter blocking, the candidate matches are examined in detail Introduction to discover true matches. This paper focuses on blocking. Record linkage is the process of matching records between There are two main goals of blocking.
    [Show full text]
  • A Hierarchical Graphical Model for Record Linkage
    A Hierarchical Graphical Model for Record Linkage Pradeep Ravikumar William W. Cohen Center for Automated Center for Automated Learning and Discovery, Learning and Discovery, School of Computer Science, School of Computer Science, Carnegie Mellon University Carnegie Mellon University [email protected] [email protected] Abstract edit distances as a general-purpose record matching scheme was proposed by Monge and Elkan [10, 9], and The task of matching co-referent records is in previous work [2, 3], we developed a toolkit of vari- known among other names as record link- ous string-distance based methods for matching entity- age. For large record-linkage problems, of- names. The AI community has focused on applying ten there is little or no labeled data available, supervised learning to the record-linkage task | for but unlabeled data shows a reasonably clear learning the parameters of string-edit distance metrics structure. For such problems, unsupervised [14, 1] and combining the results of di®erent distance or semi-supervised methods are preferable to functions [15, 4, 1]. More recently, probabilistic object supervised methods. In this paper, we de- identi¯cation methods have been adapted to matching scribe a hierarchical graphical model frame- tasks [12]. In statistics, a long line of research has been work for the record-linkage problem in an un- conducted in probabilistic record linkage, largely based supervised setting. In addition to propos- on the seminal paper by Fellegi and Sunter [5]. ing new methods, we also cast existing un- In this paper, we follow the Fellegi-Sunter approach of supervised probabilistic record-linkage meth- treating the record-linkage problem as a classi¯cation ods in this framework.
    [Show full text]
  • An Introduction to Probabilistic Record Linkage with a Focus on Linkage Processing for WTC Registries
    International Journal of Environmental Research and Public Health Review An Introduction to Probabilistic Record Linkage with a Focus on Linkage Processing for WTC Registries Jana Asher 1,* , Dean Resnick 2, Jennifer Brite 3, Robert Brackbill 3 and James Cone 3 1 Department of Mathematics and Statistics, Slippery Rock University, Slippery Rock, PA 16057, USA 2 National Opinion Research Center, Boston, MA 02114, USA; [email protected] 3 Division of Epidemiology, New York City Department of Health and Mental Hygiene, World Trade Center Health Registry, New York, NY 11101, USA; [email protected] (J.B.); [email protected] (R.B.); [email protected] (J.C.) * Correspondence: [email protected]; Tel.: +1-724-738-2508 Received: 3 August 2020; Accepted: 17 September 2020; Published: 22 September 2020 Abstract: Since its post-World War II inception, the science of record linkage has grown exponentially and is used across industrial, governmental, and academic agencies. The academic fields that rely on record linkage are diverse, ranging from history to public health to demography. In this paper, we introduce the different types of data linkage and give a historical context to their development. We then introduce the three types of underlying models for probabilistic record linkage: Fellegi-Sunter-based methods, machine learning methods, and Bayesian methods. Practical considerations, such as data standardization and privacy concerns, are then discussed. Finally, recommendations are given for organizations developing or maintaining record linkage programs, with an emphasis on organizations measuring long-term complications of disasters, such as 9/11. Keywords: epidemiology; disaster epidemiology; data matching; record linkage; probabilistic record linkage; interagency cooperation; 9/11 health 1.
    [Show full text]
  • Record Linkage Using Probabilistic Methods and Data Mining Techniques
    ISSN 2039-2117 (online) Mediterranean Journal of Social Sciences Vol 8 No 3 ISSN 2039-9340 (print) MCSER Publishing, Rome-Italy May 2017 Record Linkage using Probabilistic Methods and Data Mining Techniques Ogerta Elezaj Faculty of Economy, University of Tirana Gloria Tuxhari Faculty of Economy, University of Tirana Doi:10.5901/mjss.2017.v8n3p203 Abstract Nowadays corporations and organizations acquire large amounts of information daily which is stored in many large databases (DB). These databases mostly are heterogeneous and the data are represented differently. Data in these DB may simply be inaccurate and there is a need to clean these DB. The record linkage process is considered to be part of the data cleaning phase when working with big scale surveys considered as a data mining step. Record linkage is an important process in data integration, which consists in finding duplication records and finding matched records too. This process can be divided in two main steps Exact Record Linkage, which founds all the exact matches between two records and Probabilistic Record Linkage, which matches records that are not exactly equal but have a high probability of being equal. In recent years, the record linkage becomes an important process in data mining task. As the databases are becoming more and more complex, finding matching records is a crucial task. Comparing each possible pair of records in large DB is impossible via manual/automatic procedures. Therefore, special algorithms (blocking methods) have to be used to reduce computational complexity of comparison space among records. The paper will discuss the deterministic and probabilistic methods used for record linkage.
    [Show full text]