MetaindeX [1] is an open-source [8], online application which o↵ers an opportunity to explore and study linked archives. In this article we will focus on user interface features for loading data, displaying and studying those links within a set of documents from French Archives Nationales.
1.1 1.2
Documents and Links Modelisation
NoSQL Database MetaindeX tool relies on ElasticSearch NoSQL database
[
International (CC BY 4.0). Links modelisation Links are represented as a ’string’ field in database, containing identifiers of target documents to point at. As an example, if a catalog contains 3 documents with ID ”doc001”,”doc002”,”doc003” and we want document ”doc002” to reference documents ”doc001” and ”doc003”, then ”doc002” will contain a field (which we will call ’doclink’ for this example) with text ”doc001,doc003” as a value. Interpretation of this string is then done by application when needed.
An additional ’weight’ information can be associated to each link by adding a sux ’: <weight>’ to the referenced id. Following our previous example, value ”doc001:4,doc003” for our ’doclink’ field would then mean a link to ”doc001” with a weight of 4 and a link to ”doc003” with implicit weight of 1. This convention can be directly used in CSV files, where a column called ’doclink’ for our example would contain, for each line, IDs and weight of documents to point at.
Actual meaning of links weight is left to user interpretation. As an example, if the link is between two ”cities”, it could represent some amount of commercial exchange, while if it is between two artists, it could represent the number of peaces of art they own from each other. 2
Preparing, Uploading and Exploring Data and Links
From original Excel files have been extracted a list of unique persons on the one
hand and a list of unique archives on the other hand, as two separate CSV files
(about 4400 individuals and 2800 archives), with the help of OpenRefine tool [
Uploading contents into the server is done by dragging CSV files over the catalog contents in the user interface. MetaindeX CSV-import module allows then user to map CSV columns to new or existing fields in the catalog.
Once loaded, documents are represented as cards, each card being a single document, i.e. a single line from our input CSV files (in our case either an archive or a person). Each card can be expanded to see or edit document’s fields as illustrated on Figure 1.
We can also notice on Figure 1 that links are resolved to get a more userfriendly summary of corresponding document, rather than simply IDs list. Also jumping from one document to another, following the links, is possible by clicking on them, allowing user to navigate through those connections.
Lucene query syntax [
At last, though its usage and possibilities are out of scope of this
demonstration, we can precise that a Kibana [
Detailed usage of Gephi is out of the scope of this article, but once GEXF file generated from our data and loaded in Gephi, with only few settings we can already identify some clusters within professional and personal relationships among persons, as shown on Figure 2 (each grey dot represents a person, names have been hidden for better readability of the networks).
We can see there some clusters already well identified, where several links seem to converge around same groups of individuals. Those links were created each time two persons were found to have a professional relationship (for example master and student), seen in green, or personal relationship in pink (mariage witness, siblings, etc.). That information could help the researcher to get a better vision of social and professional relationships over his corpus, and maybe interpret with better accuracy historical facts he could find on archives contents. Since bigger graphs might be more dicult to read, MetaindeX o↵ers also a specific algorithm to generate simplified aggregated graphs. This way, nodes and links are gathered up following values of a specific field. This ”group-by” feature allows to get much more readable graphs directly focused on topic the researcher is interested in.
Following our example, we can group persons by parish they live in, which means that all persons having a ”parish” field with a similar value will be grouped within a single node, and their respective links will also be aggregated to this node.
Figure 3 shows as a result all parishes found in the corpus, their size depending on amount of individuals registered as living there, while links thickness being based on amount of links that all individuals from given parish have with individuals from another parish. On that graph, we can see that parishes SaintGermain-Le-Vieux and Saint-M´ed´eric (top right) seems to have quite numerous professional relationships (thicker link) despite their smaller amount of persons recorded to live in (smaller nodes size) from our corpus. That could maybe lead the researcher to a new approach or hints to understand social relationships of this community. 4
Conclusion Demonstration has been done that, starting from a consistent set of data, MetaindeX loads quickly and eciently the corpus, let us explore contents with advanced queries and links navigation. GEXF export module, with both basic and group-by algorithms, let us easily load our data in Gephi for advanced graph exploration. If phase of data preparation remains a major step requiring sometimes some coding skills, the proposed environment made of MetaindeX and Gephi tools, o↵ers a ready-to-use, coding-free, quick and ecient ecosystem to explore a set of linked archives for the researcher.