Life Trajectories: Semantic Enrichment and Statistical Exploration – TRAVERSEES

Biographical surveys, which are based on the collection by questionnaire of the events that mark the life course of individuals, have gradually established themselves in the social sciences and humanities as a methodological approach for explaining behaviour within populations on the basis of the individual biographical data collected. However, reconciling quantitative analysis with a qualitative understanding of life histories remains a challenge. The TRAVERSÉES project aims, through a co-constructed approach between SHS and Digital Sciences, to revisit the analysis of biographical surveys by exploiting the combined contributions of recent Artificial Intelligence (AI) approaches: the Semantic Web, Deep Learning, Large Language Models (LLM) and generative AI. An initial aim of the project is to design and implement a robust and flexible digital object, called a Semantised Life Trajectory (SLT), which will capture all the complexity intrinsic to individual life courses. The data from a biographical survey will be integrated into a Knowledge Graph (KG) structured by an ontology dedicated to the representation of SLTs and their various constituent elements (thematic trajectories, episodes, events, etc.). KGs integrate all knowledge in the form of an elementary RDF triplet (subject, predicate, object) and can be interconnected. This principle underpins the advent of the Linked Open Data (LOD) Cloud, which links together hundreds of KGs, open and linked data sets that are accessible, referencable and searchable. Once constructed, the KG of a biographical survey can be extended and completed with relevant contextual data available in the LOD Cloud. This enrichment phase will be carried out interactively using a chabot, a conversational agent linked to an LLM, which will be called upon for its ability to transform queries, formulated in natural language by an SHS researcher, into SPARQL, the KG interrogation language. These same capabilities will be exploited, not only for natural language interrogation of biographical data collected and/or enriched by a survey KG, but also for invoking the statistical analysis methods developed by the TRAVERSÉES project. In particular, to meet the specific requirements of SLT data, existing sequence analysis and neural network (Deep Learning) approaches will be adapted. A multi-level analysis will enable us to study whether the enrichment data allows us to observe certain contextual effects on life courses. Visualisation components dedicated to SLT data and to the results of statistical analysis methods will also be developed. To validate all of the project's proposals and achievements, the 3B national biographical survey dataset, which is a benchmark in the field, will be used. The TRAVERSÉES project will make a methodological and operational contribution to understanding social dynamics based on individual life histories. Through the proposed approach, SHS researchers will be guided through each of the constituent stages of a fully-equipped processing chain, ensuring the modelling, representation and exploitation (through enrichment, interrogation, statistical analysis and visualisation) of data from a biographical survey, gathered and structured within a SLT knowledge graph. The multi-disciplinary consortium involved in the TRAVERSÉES project is made up of the Grenoble Computer Science Laboratory, which is leading the project, the Espaces et Sociétés Laboratory (Sociology and Demography, Rennes), the TIMC Laboratory (Statistics, Grenoble) and Units 12 and SMS of the Institut National d'Études Démographiques (Paris).