Job offer

This PhD thesis is hosted by the I3S laboratory of CNRS at Université Côte d’Azur. More specifically, the PhD candidate will join the Wimmics team (Inria, Université Côte d’Azur, CNRS, I3S), which conducts research at the intersection of Artificial Intelligence and the Web, with a particular focus on knowledge graphs, semantic data representation and processing, and AI methods for Web data.

This PhD thesis is part of the ANR JCJC DIVA project, which investigates how user activity in interactive visualization systems can be captured, represented, and exploited to better understand and, ultimately, support users’ analytical processes. The project builds on the observation that users’ interactions with visualizations provide rich information about their analytical processes, strategies, and goals, while existing approaches often rely on ad hoc representations that limit interoperability, reuse, and automated reasoning [1], [2]. DIVA therefore aims to develop semantic and neuro-symbolic approaches for representing user activity and identifying analytical behaviours from their interactions with the system, biometric data, and verbalizations of their reasoning processes.

References [1] K. Xu, A. Ottley, C. Walchshofer, M. Streit, R. Chang, and J. Wenskovitch, “Survey on the Analysis of User Interactions and Visualization Provenance,” Comput. Graph. Forum, vol. 39, no. 3, pp. 757-783, 2020. [2] F. Yanez, C. Conati, A. Ottley, and C. Nobre, “The State of the Art in User-Adaptive Visualizations,” Comput. Graph. Forum, vol. 44, Dec. 2024, doi: 10.1111/cgf.15271.

In the context of the ANR DIVA project, this PhD thesis will focus on the semantic modelling and knowledge graph representation of users’ analytical activity, as well as on the investigation of neuro-symbolic approaches for learning and reasoning over these representations. The objective is to lay the foundations for behavior prediction and, ultimately, for the development of explainable and controllable visualization recommendation and adaptation mechanisms.

Despite significant work on capturing and visualizing user activity traces, existing representations are generally proprietary or ad hoc [1], which limits automated reasoning, interoperability, and long-term reuse. Knowledge graphs (KGs), based on Semantic Web standards such as RDF, RDFS, and OWL, offer a promising alternative for representing user activity as semantically rich linked data [3]. They enable the structuring and integration of heterogeneous activity data while providing standardized mechanisms for querying and reasoning. Such representations therefore provide a relevant basis for applying AI methods to identify analytical behaviors and predict their evolution, while maintaining explicit and interpretable knowledge representations.

Building on previous work (e.g., UI2Ont [4] and DIVA [5]), as well as open datasets on user activity [6], this PhD thesis aims to develop semantic models and neuro-symbolic approaches for representing and analysing user activity. It is structured around the following objectives:

• Semantic modelling of user activity: Develop a unified framework, based on RDF vocabularies, for representing multimodal user activity as KGs, integrating and structuring heterogeneous data generated during visual exploration, including user interactions, eye tracking, and verbalizations.

• Discovery and prediction of exploratory user behavior: Investigate neuro-symbolic methods that exploit KGs to discover and predict exploratory behaviors, combining machine learning and symbolic reasoning to identify interpretable behavioral patterns.

• Dynamic refinement and validation of behavioral models: Investigate incremental methods for refining behavioral models and rules during exploration, and evaluate their quality, interpretability, and predictive capabilities in different usage scenarios.

Work plan: The thesis will begin with a state-of-the-art review, followed by the development and validation of the semantic model and knowledge graph construction pipelines. The work will then focus on the discovery and prediction of exploratory behaviors using neuro-symbolic methods, followed by their dynamic and incremental refinement. The proposed approaches will be evaluated on real-world data and will result in scientific publications and open-source contributions.

The expected deliverables are:

• L1. Semantic model and extension of DIVA-O for multimodal user activity.
• L2. Knowledge graph construction and enrichment pipeline.
• L3. Neuro-symbolic methods for the discovery and prediction of exploratory behaviours.
• L4. Methods for dynamic refinement of behavioural models.
• L5. Experimental evaluation of the proposed models and methods.
• L6. Scientific publications and open-source resources.

Bibliographic references

[3] H. Paulheim and F. Probst, “A Formal Ontology on User Interfaces - Yet Another User Interface Description Language?,” 2011. [4] H. Paulheim and F. Probst, “UI2Ont-A Formal Ontology on User Interfaces and Interactions,” in Semantic Models for Adaptive Interactive Systems, T. Hussein, H. Paulheim, S. Lukosch, J. Ziegler, and G. Calvary, Eds., London: Springer, 2013, pp. 1-24. doi: 10.1007/978-1-4471-5301-6_1. [5] A. Menin and C. Faron, “DIVA: An Ontology-based Approach to Model User Activity within Visualization Systems,” in SAC 2026 - 41st ACM/SIGAPP Symposium on Applied Computing, Thessaloniki, Greece, Mar. 2026. doi: 10.1145/3748522.3779893. [6] S. Mohseni, A. Pachuilo, E. H. Nirjhar, R. Linder, A. Pena, and E. D. Ragan, “Analytic Provenance Datasets: A Data Repository of Human Analysis Activity and Interaction Logs,” Jan. 16, 2018, arXiv: arXiv:1801.05076. doi: 10.48550/arXiv.1801.05076.

Research Field Computer science

Education Level PhD or equivalent

Research Field Mathematics

Education Level PhD or equivalent

Languages FRENCH

Level Basic

Research Field Computer science

Years of Research Experience 1 - 4

Research Field Mathematics » Algorithms

Years of Research Experience 1 - 4