Assisted influence network entry
Once this task of familiarization with the collective categorization was finished, we developed a prototype tool to model the influence relationships between documents, actors and concepts. The goal is to assist researchers in the expression of their perspectives so that these are interoperable with the standards of Open Linked Data (DBpedia, data.bnf, Google Knowledge Graph, ISNI, etc.). In this first version, we were not seeking to automatically extract interpretations of documents by using automatic language processing (ALP) algorithms; rather, we were manually editing arguments in the form of a graph. During a second phase, we will work on the exploitation of these graphs to simulate disputes that we will present these to the researchers for validation. We aim to set up a learning cycle between the researchers and the algorithms where the perspicacity of the former is confronted with the computational reason [PAR 16] of the latter.
We used the choices made by the researchers to determine the categories that interest them (see section 2.3.1) to create “sieves”, which are so many filters allowing the elimination of menu items that are not relevant for such and such researcher. By simplifying the menus in this way, the tool’s users build their influence networks with their own categories while keeping them interoperable with those of the other researchers. In this way, users model the influence networks by graphically editing the spatial and temporal relationships between documents, actors and concepts. In doing so, they contribute to the mapping of the economy of affects [CIT 08] in a given time and space.
The actor retrieval form, for example, helps the researcher categorize an actor that he wants to add to an influence network by searching for his ISNI number [ANG 12]. After retrieving the actor’s name, the user starts the search in Data BNF to display the list of results sent by the SPARQL query. From this list, he can evaluate the person who corresponds to what he is looking for by displaying the information for each item on the list. Once his choice has been validated, the interoperability information (ISNI, VIAF, idArk, etc.) is connected to the actor who is from then on linked to the Open Linked Data.
Figure 2.12. Edit Influ: addition of an actor reference
This same form also uses the Google Knowledge Graph API to find the occurrences in DBpedia that correspond to the request that the research expressed in the form of a chain of characters. All that remains to be done then is to select the line in the list that corresponds to the sought after person to automatically obtain those long-lasting, interoperable references.
Other forms are available to recover the information concerning a conceptual reference. In this case, we ask the DataBNF service to display a list of keywords from the Rameau reference database. The researcher chooses the term from the list that he wishes to add to his reference database or whose relationships he wants to explore in Rameau. For this second need, a semantic cartography in the form of a Sankey diagram allows him to browse through the concepts to find a connected term, be it more precise or more generic. This system of choice through browsing can take place from a reference database other than Rameau, for example, the SKOS reference source made available by UNESCO or even the knowledge object database available in the SYMOGIH project.
Figure 2.13. Edit Influ: addition of a conceptual reference
A form is specially dedicated to document research. It serves to find the works used by the researcher in the BNF catalogue. It also provides the possibility to divide the document into as many fragments as necessary to model an interpretation of such and such line of a work or such and such part of a table, for example.
Figure 2.14. Edit Influ: addition of a reference to a document
The three object types (documents, actors, concepts) that make up the generic modeling paradigm are compared to a fourth object type: relationships. For this type, we have also developed a specific form that particularly allows the definition of spatial-temporal references to the relationships that exist between the other objects. To help the researcher, a map is made available to translate an address into geographic coordinates or even the simple manipulation of a marker on the map.
Figure 2.15. Edit Influ: addition of a relationship
We are continuing the development of this web application with the Paragraphe laboratory’s internal resources and the help of researchers from the Biolographes program. All of the computer sources used are available on GitHub in order to open the project to other participants and create a community allowing the development of new functionalities as well as user tests with interested researchers through influence network modeling.
-  https://developers.google.com/knowledge-graph/
-  For more information, see http://rameau.bnf.fr/
-  See https://en.wikipedia.org/wiki/Sankey_diagram
-  See http://skos.um.es/unescothes/?l=en
-  For an explanation of the project, see http://symogih.org/
-  http://www.paragraphe.info/
-  https://github.com/samszo/jardindesconnaissances/tree/master/application/views/scripts/editinflu