Creating Onomastic Social Network Maps of Books Using Their Indexes. Case Study: ‘Papyrus: The Invention of Books in the Ancient World’

Sociometric techniques deployed in documents have mainly focused on the analysis and study of scientific articles. The emergence of tools such as VosViewer (Van Eck and Waltman 2011) or Scimat (Cobo Martín 2012) has facilitated the creation of networks of terms and authors, but these have been limited to terms that appear in identification fields: title, abstract, keywords, and references (Lima and Carlos Filho 2019). This has led to an explosion of such articles, often published in journals outside the traditional sociometric and bibliometric fields, as these tools have become widely accessible. Meanwhile, books as objects of study have been sidelined due to their perceived difficulty in processing information. Books are often not digitised, academic databases such as Web of Science or Scopus are only beginning to include books, and books are typically studied in fields with a more qualitative tradition, which have usually been distant from methodologies such as social network analysis. The ease of obtaining information from Scopus and Web of Science, as well as the existence of software that automates the creation of networks and their analysis, has facilitated their creation to such an extent that there has been a boom in network analysis applied to articles. Nevertheless, books have specific fields, such as author indexes, term indexes, and so forth, that can be used to create social network maps from their contents with relative ease. It is uncommon for books to be employed as a means of creating social network maps based on the text in its entirety. The utilisation of indices (e.g., onomastic index) constitutes a valuable resource for the structured analysis of full texts, obviating the necessity for the time-consuming process of scanning the entire content. This would facilitate the acquisition of information that has not yet been subjected to analysis, which is beneficial for addressing a range of conceptual, instrumental, and applied issues. Gorziza et al. (2021) highlight that the application of these techniques to the study of books has been less frequent, partly due to the inherent difficulties in processing their information, which has traditionally not been digitised, and their focus is often more qualitative, in addition to copyright issues. Nevertheless, books remain the primary vehicle for scientific communication in the humanities (Thompson 2005) and one of the conventional means of disseminating knowledge, as well as the traditional medium for academic essays, both specialised and those intended for the general public. Despite the absence of keywords or abstracts in books (although their catalogue records do include these), they possess highly informative indexing fields, such as indices, which could facilitate the automated or semi-automated creation of social networks. The objective of this article is to present a methodological approach for utilising book onomastic indexes to construct social networks. This approach enables the application of sociometry to the content analysis of monographic works. The teleological purpose of the work is that this technique can be incorporated and used by other researchers in the analysis of textual corpora. In the context of the humanities and social sciences, social network analysis (SNA) represents a structuralist response to reductionist paradigms. Luis Sanz observes that “social network analysis is a comprehensive intellectual approach to identifying social structures resulting from the diversity of forms of relationships, but it is also a set of methods and techniques” (Sanz Menéndez 2003). The application of SNA in the humanities has been less prevalent than in other fields. It is typically employed to elucidate and examine the fundamental structures embedded in works, including character analysis, lexical analysis, and the identification of patterns within cultural, social, or historical orders. Perhaps the most intriguing aspect is that SNA in the humanities enables researchers to address substantial quantities of intricate information from a structural vantage point, offering a novel perspective through which to elucidate the interconnections between the elements under examination and to discern the existence of groups and forms that may not have been entirely apparent in a natural manner through these relationships. In this regard, there are authors who have employed social network analysis (SNA) to examine the structure of literary narratives. One of the pioneers in this field has been Franco Moretti, whose approach, known as ‘distant reading’, involves the analysis of large sets of textual data with a view to identifying patterns and relationships between words and concepts. In his article, ‘Network Theory, Plot Analysis’, Moretti (2011) demonstrates how SNA can facilitate the discovery of novel insights into narrative structure. Matthew Jockers is a prominent figure in the field of digital literature, where he has employed social network analysis (SNA) to examine the structure and interconnections of words and themes within extensive textual datasets. His book, entitled Macroanalysis, Digital Methods and Literary History (Jockers 2013), is regarded as a foundational text in this field (Jänicke et al. 2017). Another prominent researcher in this domain is Ted Underwood, who has employed machine learning methods and social network analysis to examine extensive literary corpora. His research encompasses the examination of the evolution of words and themes in literature over time, offering a longitudinal perspective on literary trends. His work, Distant Horizons: Digital Evidence and Literary Change (Underwood 2019), provides a comprehensive synthesis of his contributions to the field and his vision of the potential of SNA for the evolutionary study of literature and book sciences. Previous proposals for performing networks in books involve identifying the units to be analysed in the text, using natural language processing tools that analyse and identify terms in the text of the document (Dekker et al. 2019; Calleja Ibañez and Giménez-Toledo 2024). The issue here is trying to insert techniques prepared to be used for data and science researchers into the work of humanities and social science researchers, requiring collaboration with computational researchers. Current proposals for analysis are based on the complexity of analysing the whole text and require complex techniques for obtaining and analysing it, which are alien to the humanities, with significant exceptions such as the area of digital humanities. The basic element of the proposed methodology is the utilisation of onomastic indexes to construct the network. The utilisation of indexes offers a number of advantages. Firstly, indexes constitute a descriptive field of the document and are standardised and regulated by international standards (ISO 999: 1996). As a standardised field, it facilitates the identification of all mentions, regardless of the gender and number of the term, while also disambiguating its meaning (a primary challenge for AI used in these tasks). In addition to onomastic indexes, other types of indexes may be found, such as thematica or geographical indexes in travel books or citation indexes in theological works. The diversity of types of indexes allows for the implementation of a wide range of different studies. The initial and most pivotal phase of this methodology entails the extraction of data from the index into a tabular format comprising two columns: the first indicating the highlighted term and the second specifying the page where the term or name is documented. It is of particular importance to note that each row corresponds to a single occurrence of a term. To illustrate, if a term or name is listed in the index across pages 390–393, four rows must be generated for that concept, with each row associating the term with a single page without repetition. As these are index terms, no further normalisation is required, given that they are already standardised (ISO 999: 1996). Nevertheless, in the event that this technique is employed to generate an aggregated network by combining multiple indexes, it would be advisable to ascertain the consistency across them and, if necessary, perform normalisation on the entire set. The relationships between the index terms are determined by their co-presence or co-occurrence on the same pages. It is acknowledged that the co-occurrence of two words or two authors on the same page indicates a relationship between them. It is evident that this relationship is reinforced when both terms co-occur on multiple pages. The data are processed to obtain indirect 1-mode networks. An Excel version of the index to be processed in Bibexcel can be downloaded from the following link to enable the reader to replicate the procedure (attached to the upload). If the article is approved, (this file will be uploaded to the Zenodo repository). In order to ensure that the size of the network nodes is proportional to the frequency of occurrence, a vector is generated that reflects the number of pages on which each term appears. This is achieved by a straightforward process of counting the number of occurrences of authors in the initial table and then adjusting it to the VEC format in accordance with the author coding in the NET file. Finally, if necessary—typically due to excessive network size—the network can be reduced by eliminating relationships, nodes, or both. The method of modifying the initial comprehensive network for pruning and enhancing its visibility is contingent upon the desired outcome, which may entail the elimination of elements or relationships deemed less pertinent. Network reduction is frequently employed to accentuate the most highly connected elements or the most pivotal connections, as is automatically executed by the Pathfinder algorithm (Schvaneveldt et al. 1988). In the generation of extensive undirected networks, the utilisation of force-based algorithms, in particular the Kamada-Kawai algorithm (Kamada and Kawai 1989), is advised. This algorithm was specifically designed for the visualisation of such networks. BibExcel: This tool converts the table format into .net format. The table containing the terms or persons and the pages where they appear is inputted (one row for each occurrence, with the page number in the first column and the term or name in the second, Figure 2A). A document is generated that calculates the frequency of occurrence per author (Generate Frequencies/Whole String/Short Descending, Figure 2B). Co-occurrence frequencies in networks are calculated (Analyse/Co-occurrences/Make pairs via listbox, Figure 2C). The file is generated in .net format (Mapping and Create NET files…). It should be noted that these processes can also be automated using R, Python, or ChatGPT. A .net version of the network can be downloaded from the following link to enable the reader to replicate the procedure attached to the upload. If the article is approved, this file will be uploaded to the Zenodo repository. “Papyrus: The Invention of Books in the Ancient World” is an essay on the history of the book, written by Irene Vallejo and published in 2019 in Spanish. The text was awarded the 2020 National Essay Prize (Spain) and is currently available in 32 different languages. In his assessment, Nuccio Ordine characterises the work as “a priceless testament to an affinity for the classics, books, libraries, and reading.” Furthermore, a graphic adaptation has been published in other languages. Given its references to figures with universal resonance, this work is particularly suited to exemplifying the proposed technique. The co-presence network within the pages comprises 274 actors connected through 2411 relationships (co-presences within the pages), of which 125 have values greater than 1 (5%). The resulting network exhibits a density of 2.1%, with an average degree of 10.17 (Figure 2). The strongest relationships are those between Euripides and Plato (8), Plato and Aristotle (7), and Aristotle and Alexander the Great (7). Additionally, there are four pairs with a value of 5: Martial and Ovid, Plato and Socrates, Aeschylus and Sophocles, and Alexander and Plutarch. The network was visualised using Pajek (Mrvar and Batagelj 2016) with the Kamada-Kawai algorithm, which is specifically designed for undirected networks (Figure 3). Pajek is a powerful, open-source tool for social network analysis, known for handling large networks, advanced features, and its long-standing academic use. The positions were manually adjusted with the objective of optimising the distribution of nodes and relationships within the screen rectangle, with the intention of displaying all characters while maximising space utilisation. Furthermore, clusters were constructed with the objective of identifying the structural groupings within the network. The use of different colours to represent each cluster facilitates the visualisation of these groupings. This was accomplished through the utilisation of the Louvain community detection method (Blondel et al. 2008). The 10 most prominent characters in the book largely represent the protagonists of the work, along with their central tendency social metrics (Table 1). Despite Alexander the Great having the highest presence, Homer is more connected to other nodes and is closer to the network's centre of gravity, although not in terms of betweenness. The most significant actors are predominantly classical figures from Greco-Roman culture, with the exception of Jorge Luis Borges, Georges Perec, Leo Tolstoy, William Shakespeare, Juan Carlos Onetti, and Gabriel García Márquez. This plethora of Greco-Roman figures is an apt representation of a book that delves into the history of the book in classical antiquity, and it also mirrors the cultural milieu in which the author has chosen to situate the narrative. It is also a testament to the quality of the index itself. This proposed technique for generating networks is based on the use of indexes, a key element that enhances the quality and utility of books. “Every serious book of nonfiction should have an index if it is to achieve its maximum usefulness. A good index records every pertinent statement made within the body of the text. The key word here is pertinent.” (University of Chicago Press 1982, 512). The Chicago Manual of Style places great emphasis on the critical importance of indexes in non-fiction works. The addition of an index to a work of non-fiction serves to enhance its value, transforming it into an invaluable reference resource. This feature is fundamental to essays, scientific and technical texts, as it facilitates the rapid and efficient retrieval of specific information. While traditional co-presence techniques in articles rely on standardised fields such as title, keywords, abstract, or references, this technique proposes the use of indexes. It should be noted that the technique is only applicable to books that include them. Perhaps the most intriguing aspect of this network creation method is its simplicity and ease of implementation. It requires only a book that includes an exhaustive index of the terms with which one intends to generate a network. The approach to the content of the work is facilitated by the fact that, prior to this, an editorial professional, sometimes the author themselves, has identified key terms or conducted a thorough identification of the names of people mentioned in the work. This technique, as it is proposed, shares similarities with traditional co-citation networks as they are found in scholarly works. However, there are two fundamental distinctions to be noted. Firstly, the formation of network links is contingent upon the co-citation of two authors on the same page, as opposed to across the entirety of a given work. This approach yields more reliable similarity relationships. Secondly, the elements in question are extracted not from the references section, but from the index. It can therefore be concluded that this method represents a more precise approach to the generation of relationships. This would facilitate more expansive investigations of co-presence on pages through the capture of index entries. Furthermore, the current proposal permits the generation of subnetworks on the basis of identified chapter boundaries. In the case of works that provide a historical analysis of a phenomenon, these subnetworks can be used to visualise the evolution of networks. For example, they can be employed to illustrate the networks of individuals involved in the history of an institution. Similarly, networks from different works that explore a common topic could be aggregated to gain a thematic or person-based perspective. This would increase the complexity of the network, but it would also require a normalisation task between the different networks. The analysis of co-presence of individuals on book pages is markedly more precise than the co-presence analysis in article bibliographies generated by VOSviewer, which has been employed to construct author knowledge maps. VOSviewer's co-presence analysis aggregates all individuals cited in an article, recorded in the bibliography, and appearing across more than a dozen pages. However, the proposed technique restricts the presence of authors to a single page among the numerous pages that comprise a book. Furthermore, the technique enhances the richness of the data when the co-presence of authors is redundant, occurring on more than one occasion. This method for creating social networks provides an excellent introduction to sociometric techniques for students. Given its simplicity and potential, the lead author of this article has begun incorporating this method into their Social Network Analysis courses, enabling novice students to gain proficiency in data collection and the creation of their first social network. In principle, the basic proposal made would not allow for the underlying identification of authors. For example, Poe and Andersen might co-occur on a page discussing the difference in their characters. Another page could state ‘Poe's literature creates more horror characters than the other authors analysed.’ Andersen is not explicitly mentioned, but it is there in an underlying way. Those references could not be analysed. This limitation is undoubtedly a challenge for future research. It should provide a more complete and complex system without sacrificing simplicity. A fundamental aspect of creating social networks using co-presence on pages is that the generated network accurately reflects the method of books. In the event that the book in question a the network generated from its index will authors to the in the generated network as a knowledge that reflects the intended it a valuable technique for analysis. One that may a on the is that the co-presence of terms on the same page is on the of a work. different may be on the book This issue is particularly evident in format such as books, where the of co-presence is Furthermore, it is that two authors may be present within the same yet the network may to the relationship between them if the is across two different pages. In the presence of data serves to this in where networks are or not co-presence can be to include modifying the method to be more less This technique is particularly beneficial for that focus on the study of or as recorded in books, including literature, and cultural studies. of this is of particular to those in research in the field of Furthermore, the straightforward of the method indicates the of within book One potential application of this method is the generation of onomastic networks in or Social of books content would allow to Ibañez and Giménez-Toledo and explore a work through its networks. the of networks of this which and groups has the potential to for the of This can be achieved through the utilisation of word co-presence and cluster analysis, and analysis, and Jorge analysis, and of are due to the of this work, whose collaboration has been fundamental in that and in future of research. would also to of the for his work in research among the The authors no of The data that the of this study are available in Zenodo reference number