team Contracts

Acronyme Titre Resp. sc Début – fin
MEERQAT autre Représentation multimédia d’entités et systèmes de question réponse José MORENO
2020 – 2024
LAWBOT anr Apprentissage profond pour la Modélisation prédictive de la jurisprudence Gilles HUBERT
2021 – 2024
[Contract completed]
autre Modélisation de tâches complexes en recherche d’information Lynda TAMINE-LECHANI
2018 – 2023
Acronyme Titre Resp. sc Début – fin
NanoBubbles auropa Nano bubbles : how, when and why does sciences fail to correct itself ? Guillaume CABANAC
2021 – 2026
[Contract completed]
Analyse d’opinions sur les réseaux sociaux Mohand BOUGHANEM
2016 – 2019
[Contract completed]
Recherche d’Information Agrégative Contextuelle Mohand BOUGHANEM
2014 – 2018
Acronyme Titre Resp. sc Début – fin
BioMedExplore autre Allocation Doctorale : Approche d’extraction et de recherche d’information pour l’exploration et l’analyse multi-facettes de corpus de publications scientifiques en Biomédecine Gilles HUBERT
2021 – 2026
[Contract completed]
Recherche et Sociétés : Knowledge and INovation on Grain-legumeS food products Guillaume CABANAC
2021 – 2024

Current Projects

ANR LawBot

Deep Learning for Judicial Outcome Prediction


Description: LAWBOT is first, an applied research project in law, on the use of automated natural language processing techniques. The LAWBOT project aims to create an artificial case-law intelligence capable of predicting the judicial outcome for a given case, by imitating the decisions previously rendered by the courts on similar cases. LAWBOT is based on an artificial neural network for the deep learning of textual characteristics predictive of the judicial outcome.
Partners: LAMPS Université de Perpignan, LIG Universirté de Grenoble, IRIT Université de Toulouse
IRIS role in the project: Partner

ERC NanoBubbles

Description: The project focuses on how, when and why science fails to correct itself. To understand how the correction of science works or fails, the NanoBubbles project combines approaches from the natural sciences, engineering (natural language processing) and humanities and social sciences (linguistics, sociology, philosophy and history of science).
Partners: University Paris Sorbonne, Maastricht University, University of Grenoble-Alpes, Radboud University, University of Twente, IRIT, Ecole des Ponts
IRIS role in the project: Partner



Modelling Complex Search Tasks


Description: In the CoST project, we envision a shift from search engines to task completion engines by dynamically assisting users in making the optimal decisions empowering them to achieve multi-step and highly cognitive search tasks. This triggers the need for (1) more predictable and automatic models of user-system interactions and search tasks and, (2) more task-oriented information access models. The objectives envisioned in the CoST project are: (1) Identifying patterns of users’ behaviours while completing complex search tasks. Our aim here is to discover behavioural regularities across users and relate them through clustering techniques that could explain the nature of the involved task; (2) learning explicit and structured representations of complex search tasks, based on those behavioural patterns. Our objective here is to capture the relationships and the dependencies between task stages, i.e., the overall structure of tasks; (3) modelling task-driven IR by relating document relevance to task completion. The driving idea here is to leverage from the search patterns on the one hand and the structure of tasks on the other hand, to establish possible user actions and rewards corresponding to the accomplishment of the task. The main scientific rupture intended in the CoST project is the definition of theoretical foundations of task-based IR, a radically new IR approach
Keywords: Information retrieval, deep learning, reinforcement learning, Task, interaction
Partners: IRIT, LIP6, LIG, CLLE
IRIS role in the project: Coordinator


MultimEdia Entity Representation and Question Answering Tasks


Description: We propose to tackle the problem of ambiguities of visual and textual content by learning then combining their representations. As a final use case, we propose to solve a new scientific task, namely Multimedia Question Answering, that requires to rely on three different sources of information to answer a (textual) question with regard to visual data as well as an external knowledge base containing millions of unique entities, each being represented by textual and visual content as well as some links to other entities. In practice, we focus on four types of entities, namely the persons, the organizations, the geographical points of interest and the objects. Achieving such an objective requires to progress on the disambiguation of each modality with respect to the other and the knowledge base. We also propose to merge the representations into a common tri-modal space, in which one should determine the content to associate to an entity to adequately represent it with regard to its type. An important work will deal with the representation of a particular entity into the common space, in which one should determine the content to associate to an entity to adequately represent it. Since such an entity can be associated to several vectors, each corresponding to a data that is originally in a possible different modality, the challenge consists in defining a representation that is quite compact (for performances) while still expressive enough to reflect the potential links of the entity with a variety of other ones. The project has a potential economic impact in the fields of data intelligence, including applications in marketing, security, tourism and cultural heritage. In case of success, the output of the MEERQAT project could directly contribute to improve chatbots. During the project, the direct output will be mainly academic, that scientific article with the corresponding material to reproduce experiments. We also plan to release a new benchmark for the proposed task, in the context of an international evaluation campaign.
Keywords: knowledge base, multimedia content, question answering
Partners: LIST, IRIT, Inria Rennes Bretagne – Atlantique, LIMSI
IRIS role in the project: Partner

PNR-EST ToxMining

Approche de toxicologie prédictive pour la priorisation de l’évaluation des agents chimiques


Description: ToxMining : une stratégie d’I.A. (Intelligence Augmentée) pour exploiter les données et accompagner l’évaluation. La prise de décision lors de l’évaluation d’un composé ne peut se faire sans la contribution et l’expertise des évaluateurs. Dans ce projet, l’approche informatique vient en complément de cette expertise en donnant des pistes de priorisation des composés. Les développements en Intelligence Artificielle visent essentiellement à rendre des machines autonomes pour la résolution de tâches particulières. L’Intelligence Augmentée quant à elle est une branche de l’Intelligence Artificielle centrée sur l’interaction homme-machine, ayant pour but d’assister un expert dans son raisonnement en exploitant les capacités de traitement d’information à large échelle des machines. ToxMining adoptera donc une approche d’Intelligence Augmentée pour venir en appui de la prise de décision dans le processus d’évaluation du risque en exploitant la masse d’informations disponible en toxicologie.
Keywords: Towicologie, Métabolomite, Text Mining, Intelligence Augmentée
Partners: ToxAlim (INRA Toulous, IRIT, INSERM Rennes
IRIS role in the project: Partner

Some Past Projects