THE SOCIAL NETWORK OF the evaluation of many types of knowl- It furthermore requires that the net- DANTE’S INFERNO edge regarding both the structure of the work building procedure is able to pro- Amedeo Cappelli, Michele Coscia, text as well as the general context the duce structured results consisting of Fosca Giannotti, Dino Pedreschi, text has been generated. The context interacting entities in multi-relational Salvo Rinzivillo, KDDLab ISTI-CNR includes the whole set of historical, po- network form, computed with respect to Pisa, PI, Italy litical, social, philosophical and cultural all available data, by far surpassing what E-mail: experiences of Dante, which have been can be held in the memory of a human explicitly or implicitly involved in the scholar. In addition it is clear that the Submitted: April 29, 2010 production of his Commedia. latent knowledge emerging from the We adopt both analytical and visual analysis of the complex informative Abstract methodologies suitable to obtain a rich structure of the text is incomplete. Many Our work aims to approach the phenomenon of representation of the textual content, preconditions to the interpretation of the culture through the development of new methods and more powerful tools to capture the content of extracting and depicting what we call a Commedia are external to Dante’s text. It digitally stored literary material. As our test bed we multidimensional social network. A so- is necessary to take the knowledge on choose Dante’s characters of al di là, a domain cial network is a structure in which two Dante and his time into consideration for consisting in a set of data and relations complex enough to sharpen our tools. The methods of inves- social entities are connected by links a full interpretation. Our method tries to tigation we intent to realize should help scholars of according to one particular criterion [1]. consider both the structure of the text a literary text to concretize part of his interpretative In a multidimensional, or multi- and part of the external knowledge in a intentions. For this purpose, we will adopt advanced methodologies suitable to obtain a rich representa- relational, network the links are typed two level structure. Doing so, new com- tion, expressing the multi-relational network inher- according to particular disjoint criteria. plex paths of interacting concepts ent in the text. emerge naturally, providing different Keywords: Complex Networks, Dante, Multirela- System Specifications points of view towards the literary data. tion, Divina Commedia, Ontology, Inferno, Visual Analytics A system able to analyze and interpret a To this aim, we represent the text as a text in a relational way, must adapt itself multi-dimensional or multi-relational A literary text is a complex organism in to scholarly behavior, by expressing network, including at different levels, which many logical and knowledge enti- informative intentions regarding many the literary context together with a suit- ties interact both in an implicit and in an aspects of a text. Consequently, such a able representation of the inherent con- explicit way. In particular, new methods system must exhibit the necessary ex- text – historical, social, political, etc. and more powerful tools are needed to pressive power able to analyze and un- Although the present paper focuses on capture the content of digitally stored derstand complex multi-relational social examples, the same analysis can literary material. We do believe that problems a scholar may formulate. This be performed including any or all other complex network analysis and visualiza- may require the retrieving and the repre- dimensions. tion can provide some of these methods. sentation of a vast set of interacting con- For demonstration, we chose as case cepts in the framework. And it may lead Building the network study Dante’s characters of al di là, a to a global representation of micro and We build the respective network using domain consisting of a set of data and macro phenomena which actually repre- both the dialog relation (i.e. who talks to relations so complex to be adequate to sent the inner structures of a text – pre- whom) and several further relations match the above requirements. It is well sumably chosen by an author based on knowledge about external in- known among Dante scholars that the consciously, or on the basis of external formation used by Dante. We end up interpretation of the Commedia requires social constraints. with a network in which nodes are characters, connected if they speak to Fig. 1. Visualization of the Inferno social network, along with two conceptual ego networks, extracted from the main network and ontology data.
ghibellino francesca da rimini politico guelfo mosca dei lamberti conte ugolino alessio interminelli brunetto latini caifas camicione dei pazzi catalana ano deie maalavolti aletto megera venedico caccianemico medioevo guido guerra tegghiaio aldobrandi cavalcante pier da medicina tesifone storico niccol� iii ciacco loderingo degli andalo' alichino niccolo'nicniniccccocololo'o iiii farfarello rubicante iacopo rusticucci daal scarmiglione libicocco bertrabertberertratraam bornibob io dante farinata capocchicapoccpocchio fratelli alberti virgilio graffiacane cagnazzo paopaolp lo malal tessta bocca degli abati sinoni e un usuraio maestro adamo buoso di dovera calcabrina ciampololo pierier dellaellllalaa vignigigng a navarresnav e ciriatto diavoldiavoli guidodo dda montefeltro gerione draghignazzo barbariccib aric a frate alberiga o malacoda fleglegiias capanean o caroncaroonte griffolino d'arezzo dannati in arrivo ciampolo navarrese filippo argenti nesso cagnazzo giacomo da sant'andrea vanni fucci virgiligi io suicidic a fiorentino alichino draghignazzdr ghign zzo maometto ulissulis e centauri libicocco loderingo degli andalo' chirone malacodm a catalano dei malavolti scarmiglione minosse farfarellfar arello rubicante diavoli guardiani pluto messo celeste barbariccia graffiacane minotauro diavoli calcabrina ciriatto dante dante dante dante
virgilio virgilig io
Fig. 2. Temporal evolution of the Inferno social network within the narrative timeline.
acters, connected if they speak to each torical” in the ontology using the link provides a browsing feature able to focus other, and concepts connected to relevant “character type”, to “Medieval” in using the attention of the scholar on a particu- characters. These relations were ex- the “historical period” link, and to lar topic. With this system, one could tracted from a manually annotated ver- “Guelph” using the link type “political address some questions like: Why does sion of the Inferno, according to a faction”, etc. Dante use a dialectal lexicon for a par- process explained in [2] and [3], where With this additional information we ticular character?, Are Guelphs and Ghi- relevant information about the speaking enable users, who are interested in study- bellines randomly distributed in the text characters is collected and linked to the ing the interplay among several different or are there some particular patterns in dialog by a domain expert, using the concepts and characters, to navigate and their appearances?, etc. social, political, historical, mythological explore such relations. Initially, we visu- Furthermore, we are confident that and linguistic contexts. alize the main social network without the some non-trivial analytical results could In the main social network the entities ontology information. Then the user can also be achieved. Consider Fig. 2, where are the characters of the Inferno. A select one particular character and focus according to the visible the evolution one weight is assigned to link according to on all the concepts related to him, re- can say that the network exhibits a “win- the number of verses in which one char- stricting the visualization to immediately ner takes all” effect – very similar to the acter talks to another. The whole social relevant concepts and further characters well known preferential attachment network of the dialogues is represented sharing the very same concepts. Since model of the Web [5]. The two main in the left part of Fig 1. these networks are centered on a particu- hubs, Dante and Virgilio, and their dense There are two additional information lar entity, we call them “ego networks” communication links, reflect the central levels attached to this social network. [4]. The process is depicted symbolically role that Dante has reserved for Virgilio, First, we have “temporally annotated” in Fig 1, where in the right part we show representing the magister that drives the connections according to the part of two possible results of two different se- Dante through the al di là. the Inferno in which the dialogue ap- lections: the ego network of Malacoda pears. Subsequently it is possible to and Catalano dei Malavolti. References and Notes visually grasp the evolution of the flow * This paper was presented as a contributed talk at Arts | Humanities | Complex Networks – a of information present in the Inferno. Results Leonardo satellite symposium at NetSci2010. See Fig. 2 gives an impression of this dy- As stated in the system specification namic evolution, showing four snapshots section, the main focus of our work is to 1. M. E. J. Newman, The structure and function of corresponding to four of the original create a novel point of view on the inter- complex networks. (SIAM Review 45, 167-256, divisions in the Canti – VI, X, XVI, and acting concepts of a literary text. We 2003). XXII. have described a system able to create an 2. A. Cappelli, M. N. Catarsi, P. Michelassi, L. Second, we have linked each character Moretti, M. Baglioni, F. Turini, M. Tavoni, Knowl- immersive knowledge environment. In- edge mining and discovery for searching in literary to concepts in an ontology. For example, side this environment, the scholar is sur- texts. (Proceedings of the Third International Con- consider the character Catalano dei rounded by a complex, multi- ference on Language Resources and Evaluation, Las Palmas, 2002). Malavolti, an historical character that is a dimensional relation space, defined by medieval guelph politician present in the all the concepts the author included, 3. Miriam Baglioni, Mirco Nanni, Emiliano Gio- vannetti, Mining Literary Texts by Using Domain Cerchio chapter VIII. Therefore, the implicitly and explicitly in his work, as Ontologies. (Proceedings of the Knowledge Discov- character node “Catalano dei Malavolti“ far as they are made explicit by the cura- ery and Ontologies workshop in ECML/PKDD, is connected to the concept node “His- tors who annotated the text. The system Pisa, 2004). 4. Linton C. Freeman, Centered graphs and the construction of ego networks. (Mathematical Social Fig. 3. The workflow of our work: from the plain text to the complex network. Sciences Volume 3, 1982). 5. Albert-Laszlo Barabasi, Reka Albert, Emergence of scaling in random networks, (Science 286, 509, 1999).