Missions

The RADID SUMINO project (sponsored by Agence Innovation Défense or AID) is part of the "User interactions: systems for automating repetitive and time-consuming tasks" theme, and provides a new response to a well-known and frequently encountered problem: multi-document management for information retrieval.

In this 24-month project, we are addressing this issue along two axes, representing the two main objectives:

(1) to create a hybrid single- and multi-document summarization tool based both on unsupervised learning and on the exploitation of an ontology of the targeted domain,

(2) to couple this tool with a search engine also based on business ontologies and with a tool for tracking topics over time (diachronic analysis).

Activities

The work on the LORIA side will involve adapting research work carried out in the SYNALP team to produce automatic knowledgeless extractive summaries, in particular adapting existing work to multi-document extractive summaries and knowledgeless short text summaries. It will also involve interfacing the extractive summarization methods obtained with ontology-based methods to produce abstractive summaries.

Finally, the work carried out should make it possible, in a second phase, to set up methods for monitoring topics over time.

Skills

Clearly expected experience in natural language processing (NLP):
- Familiarity with language models and NLP tools.

Good general experience in machine learning, particularly in unsupervised learning:
- Mastery of explanatory learning models.

Proficiency in programming languages/environments:
- Python
- C/C++

Work Context

The work will take place at LORIA and, if selected, the candidate will be fully integrated into the SYNALP team. He/she will benefit from a rich and solid research environment, in an internationally recognized team. He/she will take part in face-to-face and remote work meetings with AIRUDIT, the project partner.

He/she may also take part in project follow-up meetings with AID.

Constraints and risks

The main constraint will be to meet the deadlines imposed by the RAPID SUMINO project sponsor, AID, which will monitor the results described in the project deliverables:

R1: Implementation of an extractive, knowledge-free summarizer operating in different modes (short documents, long documents, multi-documents) (T0+12).

R2: Implementation of a process for collaborating resumers obtained with ontology-based approaches to produce abstractive summaries (T0+18). R3: Implementation of subject tracking methods (T0+24).

R3: Implementation of topic tracking methods (T0+24).

Additional Information

None.