I work on modeling and experimenting on indexing and retrieval of text and visual data.
- I focus right now on continuous evaluation of Web search engines in the Kodicare project (ANR PRCI, leader);
- A also work on explainable video retrieval;
- I also concentrate on normative documents indexing and retrieval, in collboration with Schneider Electric;
- I did focus on theoretical models for information retrieval (for instance using Jensen-Shannon divergences) , and on theoretical hypotheses and behavior of such models (like the Document-Document Matching Constraint). The question of the existence of such constraints for continuous representations (LDA, LSA, Deep learning) representations of indexing terms is also a great challenge to tackle;
- I participated to evaluation campaigns for Multimedia indexing and retrieval: NTCIR Lifelogging, CLEF Image Annotation;
- I did, with Lorraine Goeuriot, manage a task for the CLEF Tweet Contextualization Lab, dedicated to propose a way to evaluate the quality of indexing and retrieval of tweets from festivals;
- I proposed, with Yves Chiaramella and Franck Fourel, the “Fech&Browse” framework for the indexing and retrieval of structured multimedia documents. This idea inspired numerous works in this field, especially in the context of the INEX evaluation campaigns between 2003 and 2010. The “Fetch&Browse” rely on notions of specificity and exhaustivity between queries and documents;
- I worked until october 2016 on the integration of language models on graphs to describe photographs. The idea is to be able to extract features, and to represent relationships between these features in the index of images. This wok is valorized by the CNRS, in a scientific collaboration with the Globe VIP company.