Today your reporter will deliver this post live from her new temporary habitat: the laundromat. I was hoping it would never come to this, but the landlord still hasn’t installed pipes for the washing machine. I’ve been postponing this ordeal for over a month now, and my laundry basket simply can’t handle the steady heaps of clothing being catapulted its way anymore. So I stuffed everything into my backpack, trolley, Ikea bag, any big container I own basically, and dragged myself to the nearest facility, with the very deceitful name ‘Happy Wash’. Yet there is absolutely nothing spirit-lifting about sitting in a fluorescently-lit corridor lined with machines shaking and snorting as if they’re about to take off. The owners are obviously of the hippy-dippy retro sort: there’s no Wi-Fi. One might argue that the absence of all technological diversions can be seen as a blessing, as it allows me to fully unleash my literary genius and focus on the task at hand: providing you, dear reader, with entertaining and educational refreshments. But somehow I find that being surrounded by glass doors offering a window into other people’s tighty whities being swirled around oddly discomforting and distracting (though luckily not satisfying).
Today’s special is geared toward the prosopographically-minded. As we’ve been repeating for years now, you can’t do social network analysis if you don’t have a decent prosopography. It’s become kind of a mantra. You need to get your data sorted out.
Now, while working on Trismegistos is awesome, it does come with itsie bitsie teeny weeny downside (no bikini’s, sorry, Leuven doesn’t exactly have a very active beach community): although we literally have thousands and thousands of people in our database, it’s a brobdingnagian mess. Allow me to digress for just a moment to explain how TM People, our onomastic-prosopographic structure, works.
As all things in life generally do, we start from the beginning, the basics, the elementary particles which historians use to recreate the past: attestations in sources. Words to big for ya? Let me put it this way: (almost) each time a person is mentioned in a text, it is included in our database. I say ‘almost’, since I’m not going to be so bold to claim we’re infallible and haven’t missed a single thing. Processing all these names nearly cost me my sanity, but we also ended up with almost half a million attestations, so I guess it was worth it. But like I said: you can’t do SNA if you don’t have a decent prosopography. And unfortunately, a database of mere attestations is not a decent prosopography. Let me enlighten you with a simple example.
Hundreds of years from now, assuming a brave band of heroes succeeded in stopping Trump wipe out mankind completely, future historians will no doubt find plenty of documents mentioning the name ‘Yanne’. It can be as simple a scribble as ‘Yanne loves Oreo’s’. In this particular sentence and context, the term ‘Yanne’ obviously refers to me, a damsel born during the last quarter of the twentieth century, famous for her unconditional love for Oreo’s, who made an attempt at a depressing career in academia (in which she succeeded marvelously, seeing that she became unemployed after only four years) and died alone at the age of 53, half-eaten by Alsatians (where those came from, only God knows, because she couldn’t even manage to keep a cactus alive, so she knew better than to keep pets).
Lets say these future historians find a hundred Oreo-related scribbles mentioning my name. Would their futuristic network contain a hundred Yanne-nodes? Of course not! As much as I would like to master the art of bilocation sometimes, I still am, and most likely for the rest of my rapidly shortening life will be, a single, whole, undividable human being. Those one hundred attestations will be useful to generate the many, many links in the network (I like to share, even my precious Oreo’s), but on the node level, there’s just one me. Now, chances are that those future historians will have much smarter computers than we do, and will most likely be able to decide for themselves that all those references point to the same person. But let’s assume that Trump eliminates all Humanities funding, both Digital and analogue, and computers are still as forlorn without human guidance as they are now. Their PhD minions will be forced to go over each of those attestations to check whether they do indeed refer to me, and mark them as such. And repeat that for every person they want to include in the network. Not as exiting as it sounds.
But, oh, if only it were this simple! Alas, the trials a creator of prosopographies faces are usually much more challenging. Let me continue with my illuminating example. Obviously, I don’t own the rights to my name, although I would very much have liked to. Now as long as Oreo’s are at stake, everyone knows it’s about me. Those cookies are my life, they define me. But what if one day that haggard PhD student comes across a document mentioning Yanne’s secret stash of Le Parfait.
I’m so sorry, dear reader, but I feel the very urgent need here to digress, since I’m afraid that most of you are not familiar with this delicacy. Le Parfait is a lesser-known and more recently cultivated addiction of mine. I couldn’t have come up with a better name myself for this Swiss delight, brought to you in a tube. Yes, like toothpaste. But with food. Mouthwatering, deliciously creamy, succulently savory pâté. That’s what it is. Moreover, thanks to its quirky conservation mode, it is the ideal staple to survive any biological, nuclear, social media, or traditional war. Trust the Swiss to get you geared up properly for survival mode. Unfortunately, next week my dealer is leaving Belgium for, well, indefinitely, so I’ve had to start rationing my supplies. In the meantime, I’m waiting for Nestlé’s reply: if they decline my offer to become a Belgian mascot in return for a lifelong supply, I’m thinking of starting up a smuggle route. What else are those Alps for anyway?
Back to our story. Since there is no prior association between me and this tube filled with perfection, the PhD student is hesitant. This poor sod knows that I would never in my right mind forsake Oreo’s, so this must be someone else here. But in a post-aTrumpalyptic world where only canned and tubed food can thrive in the globally overheated environment, nobody could afford to be too picky about food. How can my biographers be sure that I was the Yanne with the tube? I have no doubt inspired many a proud parent who reads this blog to name their Wunderkind after me; and no doubt some of these disappointing brats will survive the tidal waves of disasters that Trump is paving the way for so eagerly these first days of his presidency.
The answer to this sticky wicket is, of course: make a network! As a diligent reader, you are now spluttering befuddledly: didn’t I just say that in order to make a proper social network, you need to clean up your data first? Indeed! But as Silke and I discovered, even improper networks can be a great help. We were planning on publishing a paper on the subject, but then life got in the way, as it usually does. Since I don’t want all that hard work to go to waste though, I came up with a brilliant plan: publish it in my upcoming book! And since the book is based on the blog, I might as well publish it here as well as a spoiler. So this will probably be the only (semi-)formal post you’ll ever see here. Yes, this means that, in a minute, you’re finally going to read a proper scientific paper. Don’t be shy. You’ll manage. We’re human, just like you. Unless you’re not. Then please don’t tell, or I’ll freak out. I’ve already been sleeping badly lately since I found out that I’m worth only 50 camels, while everyone else at the office would sell for 65 at the least. Apparently I’m too old, and my boobs are too small. Going up a cup size or two would score me at least 4 more. Now, I don’t know what breast implants cost, but according to Africa Geographic, among the Maasai, in 2013 one camel cost 36 goats or sheep, or around $700 (africageographic.com/blog/whats-a-camel-worth). And in Britain (obviously a country where camels have always been an important economic asset), a single one could even fetch £7,000 (www.dailymail.co.uk/news/article-2237967/Camels-sale-Britains-biggest-herd-market-13k-each.html)! I’m seriously considering the surgery, guys…
(And it’s kamelrechner.eu/en by the way).
Oh, a final note on the paper: I’m going to leave out the second part, because that’s going to be a very, very important chapter in Silke’s PhD thesis (no pressure, hon), and I wouldn’t want to give away any spoilers! Ok, here it goes!
The Only Formal Post on this Blog, aka Identifying Individuals through Network Visualization
The Leuven-based Trismegistos database, an interdisciplinary portal of papyrological and epigraphic sources from the ancient Mediterranean (800 BC – AD 800), currently contains over 300,000 texts and almost half a million attestations of individuals. While perhaps not exactly ‘big data’ in the eyes of many, it presents some exciting challenges regarding data management and analysis to the scholars working with it. One of the tasks we are struggling with is the disambiguation of individuals. Within papyrology there is a strong tradition of prosopographical research, yet in the past these studies focused on small data sets, such as archives, since identifying individuals across texts is a very time-consuming process, for which you need to take many variables into account (homonymy, family relations, titles, ethnics, …). It is obvious that the 400,000+ references in Trismegistos cannot be tackled in the same way. Where available, existing prosopographies, e.g. the Prosopographia Ptolemaica, have been incorporated, but these cover only a fraction of our data. Without discarding this longstanding tradition, we therefore set about to seek a method that could help us manage our data more quickly and at the same time provide us with additional arguments and thus a more solid basis to identify individuals.
While working on projects using social network analysis, we found that network visualizations offer several advantages when carrying out identifications: clear overviews, simplicity and speed. This paper is aimed at explaining how we use these visualizations for disambiguation purposes. Furthermore, we have tested the efficiency of this identification method and its applicability for different aspects of prosopography, such as individual identification and the identification of family and business relations.
With the advance of new digital techniques in the humanities, prosopographical study has made a comeback on a much larger scale. The issues and solutions raised here are not limited to data from the ancient world, however, and are therefore applicable to a wider range of prosopographical databases. Furthermore, they can be a useful addition to the way network analysis is applied to biographical data in a more “traditional” way, e.g. in the Berkeley Prosopography Services project or the China Biographical Database project, as well as for data mining purposes.
Our data is stored in the Trismegistos [TM] platform (www.trismegistos.org), which consists of multiple related databases, set up in a mixed Filemaker 12-13 environment with a single server. It was designed to incorporate the metadata for all documentary texts from Egypt dated roughly between 800 BC and AD 800. Since 2012, TM is expanding its geographical scope, starting off with texts in ‘smaller’ indigenous languages such as Etruscan, Italic, Messapian, Lepontic or even Runic, and by 2014 all Latin inscriptions were incorporated as well. Building on the Prosopographia Ptolemaica, TM People (see below) has over the past couple of years been expanded to include all references to individuals in the Egyptian texts.
The central text database [TM Texts] currently contains more than 350,000 records, mostly of papyri and ostraca (i.e. potsherds used as a writing surface), but also inscriptions, writing tablets and graffiti. All attested languages and scripts are represented, including Greek, Egyptian (including Hieroglyphic, Hieratic, Demotic and Coptic), Latin, Aramaic and Arabic. Essential metadata about the texts is made available, such as inventory numbers and publications, writing surface, language and script, date, provenance, type of document, etc. Each text is assigned a unique number, the text ID or TM number, which is used internally to link to the other databases of the platform, but is also increasingly employed to identify texts in other databases, such as the Papyrological Navigator (e.g. the publication SB 14 11714 = TM 14512 and is consultable at www.trismegistos.org/text/14512 with a link to the full text in the Papyrological Navigator at www.papyri.info).
From the text database it is possible to access information about the individuals mentioned in the texts. Two interconnected databases were set up in TM People, a prosopographical database collecting all individuals [PER], and one for the references to these individuals [REF], as people can appear more than once in the same or in different texts. Heron, for example, is known as TM Per 227434. He was already an old man when mentioned his the documents that survived: the most notorious is a petition to the governor of Egypt in AD 222 asking to be exempt from public services on account of ill health and old age (he was 70 years old by then). In total, he appears 15 times in seven different texts. Each attestation is recorded separately in the REF database. This database collects such information as the person’s name as it appears in the text, where precisely he can be found (line number), as well as titles, ethnics or status designations. Further information is extracted from the TM Texts on the one hand, e.g. the date and provenance, and from the text itself on the other, such as the person’s role in the text (i.e. is he the subject of the identification cluster, or is he merely mentioned as a patronymic).
The starting point for the prosopographical section was the digitized version of the Prosopographia Ptolemaica, which collected all individuals with a title attested in Egyptian, Greek and Latin documents with a title living in the Ptolemaic period. Since 2008, we have been entering other individuals occurring in TM Texts, not only those with a title. Different types of documents required a different approach. Much of the data needed to be entered manually: this included attestations in Greek and Latin epigraphic sources, Demotic papyrological and epigraphic material, and all Hieroglyphic and Hieratic documentation. Names in Coptic papyri were integrated with the help of Alain Delattre’s Brussels Coptic Database. The majority of our documentation, however, some 50,000 Greek papyri and ostraca, was tackled in a semi-automatic Named Entity Recognition [NER] process on the basis of the Duke Databank of Documentary Papyri [DDbDP]. This online corpus provides the Unicode XML full text of all published non-literary Greek and Latin papyri, ostraca and wooden tablets. In the first stage 343,071 references to individuals were extracted from the Greek texts dating between 332 BC and AD 500 and were imported into TM. Together with the manually entered data the prosopographical structure now contains 458,012 references to 357,978 individuals, and the core information is accessible online through the portal site at www.trismegistos.org.
Work on this prosopographical structure is not yet finished though. As the NER mark-up and extraction of individuals with their genealogical information were checked manually text per text, people could not be identified as one and the same person across documents. Since homonymy was a common practice, this crucial next step was felt to be too delicate for NER automation, and with over 350,000 PER records, this is an enormous task to undertake manually. Given the long history of the Prosopographia Ptolemaica and the more limited number of texts, most of the work for the Ptolemaic period has already been done. For the Roman and Byzantine period, however, identifications have only been carried through in light of research on specific archives or types of names (e.g. double names). For exceptionally long documents, such as the tax rolls of Karanis, in which recognition of recurring individuals within the text was not possible during the extraction process due to the immense number of names they contain, identification has partly been realized automatically for those with a complete match of their name, patronymic, papponymic and metronymic.
On top of the large amount of data, the identification of individuals is furthermore impeded by the scant background information available to us on the subjects. In some cases, a combination of titles, ethnics or family relations is suggestive. In AD 330 an Aurelia Demetria-Ammonia, daughter of Polydeukes, a former magistrate and councilor of Hermopolis in Egypt, petitions an official; an Aurelia Demetria-Ammonia, daughter of Polydeukes, former gymnasiarch and councilor of Hermopolis is mentioned in a contract from AD 328; and in a lease contract (AD 332) an Aurelia Demetria-Ammonia from Hermopolis is found. The double name, the patronymic, the father’s titles and the city leave no doubt that this is one and the same person. She appears in seventeen other texts, among which court proceedings from AD 299 where she is simply styled as Aurelia Demetria daughter of Polydeukes.
For many people, however, no information other than their names is available: there is a lack of both patronymic and title information, and the context is not very informative. These uncertain examples have so far not been identified with namesakes. Thus three people with the name Dionysios-Amois from the second century AD have been kept apart, although they might be one and the same, or they might even be the same as one of the people called Dionysios-Amois for whom a patronymic is known. Some names are so common that their bearers cannot even be identified with certainty despite patronymics or titles.
By drawing on graph visualization and network measures, we now have a helpful technique at our disposal when carrying out prosopographical identifications in Trismegistos. The idea grew when we started exploring the possibilities of network analysis for the TM data set. The analysis of personal networks requires a decent prosopography, and as such we were confronted with the shortcomings of what was available. In some preliminary case studies, we discovered that, despite the lack of background information, people could be identified on the basis of mere co-occurrence. Although TM gives an overview of all people who appear in the same text, lists like these are not very practical, especially for long texts with many attestations, and for large datasets. Comparing two or three texts may be feasible on a standard computer screen, yet even this is very time-consuming. The method presented here gives an overview of all data in a single graph, and allows to “zoom in” on specific individuals and to compare their “surroundings”, i.e. the people with whom they appear in the same texts.
To start with, the data set is exported from Filemaker as two separate CSV files: a node list and an edge list. The node list contains all background information that can help with the identification of individuals, e.g. the person’s name(s), possible titles and ethnics, his patronymic (and papponymic, metronymic, … if available), the date range and provenance of the texts in which he is attested, etc. All these characteristics are considered attributes; to avoid confusion between homonymous individuals, the unique numeric identifiers assigned to these individuals in TM are used as their ‘ID’ instead of their names. As a result of the interlock structure of TM People and TM Texts, a simple two-mode affiliation matrix of people-in-texts can be extracted from TM instantly. This edge list consists of two columns, the first containing the person IDs, the second containing the TM numbers of the texts in which those people are attested.
Since we are looking at patterns of connected individuals, our two-mode people-in-texts network needs to be converted into a one-mode people-to-people network (Table 1):
|TM Per||appears in||TM||⇨||TM Per||appears in TM 16504 together with||TM Per|
Several software packages specialized in network analysis offer a built-in conversion tool. In Gephi, the software used for this paper, the ‘Multimode Networks Transformations’ plugin serves this purpose. Once the data of this one-mode network is loaded in a network visualization tool, such as Gephi or Netdraw, the identification of individuals can start.
The project ‘Defining the Elite in Roman Egypt’ focuses on the two privileged orders of Roman Egypt, the ‘metropolites’ and ‘those of the gymnasium’. They were founded under Emperor Augustus somewhere between the end of the first century BC and the beginning of the first century AD. Registration in a metropolis, one of Egypt’s district capitals, was mandatory, and status could only be passed down to the next generation if both parents were members. To gain a better understanding of how these two groups related to each other and evolved, the first author is setting up a prosopography of their members. Status designations are barely used, which was one of the reasons to explore alternative methods for tracking down possible identifications.
Hermopolis. As a first test case, we therefore focused on a metropolis in Middle Egypt called Hermopolis, the city dedicated to the Greek god Hermes and his Egyptian counterpart, the baboon god Thoth. Seventeen texts dated between AD 50 and AD 80 were collected, containing 157 attestations of “different” individuals. All attributes that might help with the identification were added to the node list: titles, ethnics, status designations, the role the person plays in the text, and patronymics (if available of course). After conversion of the ‘people-in-texts’ network into a people-to-people network, the data was visualized in a network graph containing 157 nodes (the “individuals”), connected by 934 edges (documenting co-occurrence in the same text). It consists of seventeen unconnected components (Figure 1). Each component represents a different text. Figure 2 zooms in on a component in the upper right corner of the graph. These are the individuals that appear in P. Heid. Gr. 4 339 (TM 21117); the cluster to the left are those mentioned in P. Ryl. Gr. 2 101 a (TM 19497).
Since the names are added as labels in Gephi, we color-coded them so patterns could be discovered more easily. By highlighting a specific name, components can be scanned fairly quickly to check if other names reoccur as well. In a first instance, the name Diodotos was selected, a standard Greek name that appears in three different texts (Figure 3), without additional information. When going over the other nodes in these clusters, the rare name Metokos immediately stands out: in Figure 4 it is highlighted together with Diodotos. Each time, the name Metokos is followed by the title sitologos. Since all these texts are dated between AD 65 and 66, there is little doubt that this is one and the same person. Moreover, Metokos is the addressee in each of these texts, while they are all sent by people called Diodotos. All should clearly be identified as a single Diodotos. More clues are provided by the names Gaius Iulius Salvius (the turquoise nodes in Figure 4) and Sophos (the green nodes in Figure 4), which appear in two out of the three texts as well (and Sophos is styled a slave in both). These four people, Gaius Iulius Salvius, the slave Sophos, Diodotos the sender and Metokos the sitologos and addressee, form the center of this small-scale (and temporary) ‘business’ network. The identification is based on the co-occurrence of these people and their attributes.
The archive of Tryphon. The archive of Tryphon, a weaver from the town of Oxyrhynchos, served as a second test case. It comprises forty-three texts dated to the early Roman period. They include 487 references to 194 “individuals”. Since the main actor of this archive, Tryphon, and some of his family members (his father, mother and wife) have already been identified across most texts, this network already has a giant component centering around Tryphon. The clusters referring to individual texts are nevertheless still clearly visible (Figure 5). The problem with this archive is that it contains many homonymous people with very common names, such as Sarapion and Ammonios, so it is tricky to identify them. Focusing on the four names that recur most often (i.e. Sarapion, Ammonios, Thoonis and Theon) by highlighting them as in the Hermopolis example, we were able to identify ten people, since a combination of two or more of these names recurred in several clusters. The four names appear together in two clusters; so does the combination Sarapion and Thoonis. The combinations Sarapion + Ammonios + Theon, Sarapion + Theon, Thoonis + Ammonios, Sarapion + Ammonios and Theon + Thoonis are each attested in one cluster.
Admittedly, the data sets for these test cases are relatively small. Plowing through the database manually might eventually have led to similar results. Creating visualizations as we did here is much more time-efficient, however. It suggests avenues for the exploration of interesting combinations of names, allowing also to maintain an overview of all the relevant records. It provides an answer to the issue of scalability, as it provides opportunities for analysis of large datasets. Moreover, where traditional historical-prosopographical research is often speculative, the network approach adds criteria through measures such as centrality and distance, which are calculated on the basis of the links that are actually attested in the documentation. In the current process, the databases and network visualizations and calculations remain locked away in separate programs. Further integration of e.g. visualization techniques and network-based automated suggestions for identification into TM remain interesting prospects for the future. Experiments with probability calculations, such as the social rank inferring method tested on the cuneiform tablets of Kültepe, furthermore show great potential to improve the identification process.
We are also looking towards other projects focusing on ancient prosopography where we believe this method could both advance the identification of individuals and the study of prosopographical information as well as take advantage of larger prosopographical datasets available in collaborations such as in the SNAP:DRGN project. SNAP:DRGN connects different databases with prosopographical information and “[builds] a virtual authority list for ancient people through Linked Data collection of common information from many collaborating projects”. At the moment, SNAP:DRGN would not be a suitable source for network analysis, because of the shallowness of the shared data, but network analysis may be an interesting tool to enrich the data in the future.
And now if you’ll excuse me, my Parfait and Oreo’s are calling.
 For an introduction, see A.-L. Barabási, Linked: The New Science of Networks, Cambridge (Mass.), 2002; and T. Brughmans, ‘Facebooking the Past: A Critical Social Network Analysis Approach for Archaeology’, A. Chrysanthi et al. (eds.), Thinking Beyond the Tool: Archaeological Computing and the Interpretative Process, Oxford, 2012, 191-203.
 E.g. several projects related to King’s College London: the ‘Prosopography of the Byzantine World’ (www.pbw.kcl.ac.uk), the ‘Prosopography of Anglo-Saxon England’ (www.pase.ac.uk/index.html), and ‘People of Medieval Scotland’ (www.poms.ac.uk); ‘Mamluk Political Prosopography Project’ at Ghent University (www.mamluk.ugent.be/prosopography); and ‘People of the Founding Era’ at the University of Virginia (pfe.rotunda.upress.virginia.edu).
 See, for example, K. S. B. Keats-Rohan (ed.). Prosopography Approaches and Applications. A Handbook, Oxford, 2007.
 F. Rossi, N. Villa-Vialaneix and F. Hautefeuille, ‘Exploration of a Large Database of French Notarial Acts with Social Network Methods’, Digital Medievalist, 9 (2013) (www.digitalmedievalist.org/journal/9/villavialaneix/).
 For more information on the current coverage, see www.trismegistos.org/about_languages.
 TM also consists of further databases dealing with onomastic information, recording all names attested in the documents and their variants, and a geographical data, but these are not relevant here.
 Jut like the TM numbers, the identifiers used in the other Trismegistos databases are part of stable uri’s. Heron can therefore be consulted at www.trismegistos.org/person/227434. For more information on how to consult and cite Trismegistos, see www.trismegistos.org/about_how_to_cite.
 L. Mooren, ‘The Automatization of the Prosopographia Ptolemaica’, in I. Andorlini et al. (eds), Atti del XXII Congresso Internazionale di Papirologia (Firenze, 23-29 agosto 1998), Florence, 2001, 995-1008.
 Y. Broux, Double Names and Elite Status in Roman Egypt, Leuven, 2015.
 P. Mich. 4 223-225 (TM 11998-12000, AD 172-175).
 Stud. Pal. 2 p. 33-34 [N.N. 44], l. 1-2 (TM 20813); Stud. Pal. 20 86, l. 2 (TM 18728); and SB 14 11711 (TM 18176), l. 1.
 P. Kramer 11, l. 2 (TM 33347).
 In (social) networks, nodes (or actors) are people; they are connected by edges (also called ties or links).
 E.g. UCINET (sites.google.com/site/ucinetsoftware/home), where two-mode networks can be converted under Data > Affiliations (‘2-mode to 1-mode’); or R, where this is included in the ‘tnet’ package.
 Gephi.github.io. A tutorial for this plugin can be found here: electricarchaeology.ca/2012/04/04/converting-2-mode-with-multimodal-plugin-for-gephi.
 Y. Broux, ‘Creating a New Local Elite: The Establishment of the Metropolitan orders of Roman Egypt’, Archiv für Papyrusforschung und verwandte Gebiete, 59 (2013), 142-152.2
 P. Ryl. Gr. 2 294 descr. is not yet published; a brief description with a transcription of one line mentioning Herakleios-Kronion is given, so he cannot be connected to any other names.
 P. Lond. 3 p. 121-122 no. 1213 (TM 22837, AD 65), where it appears twice; P. Lond. 3 p. 122 no. 1214 (TM 22839, AD 66), where it appears twice; and P. Lond. 3 p. 121-122 no. 1215 (TM 22841, AD 65).
 Official in charge of the collection of taxes in grain.
 A giant component is a connected component in a network in which a significant fraction of all the nodes of the network is contained. For more information on this and other network-related concepts and measures mentioned throughout this paper, see Wasserman and Faust 2009.
 D. Bamman, A. Anderson, N. A. Smith, ‘Inferring Social Rank in an Old-Assyrian Trade Network’, arXiv:1303.2873v1 [cs.CY].
 G. Bodard et al., Standards for Networking Ancient Prosopographies: Data and Relations in Greco-Roman Names, 2014- , available: snapdrgn.net