Author name cluster

Christophe Gravier

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

5 papers

2 author rows

JMLR Journal 2025 Journal Article

Fair Text Classification via Transferable Representations

Thibaud Leteno
Michael Perrot
Charlotte Laclau
Antoine Gourru
Christophe Gravier

Group fairness is a central research topic in text classification, where reaching fair treatment between sensitive groups (e.g., women and men) remains an open challenge. We propose an approach that extends the use of the Wasserstein Dependency Measure for learning unbiased neural text classifiers. Given the challenge of distinguishing fair from unfair information in a text encoder, we draw inspiration from adversarial training by inducing independence between representations learned for the target label and those for a sensitive attribute. We further show that domain adaptation can be efficiently leveraged to remove the need for access to the sensitive attributes in the data set we cure. We provide both theoretical and empirical evidence that our approach is well-founded. [abs] [ pdf ][ bib ] [ code ] &copy JMLR 2025. ( edit, beta )

PDF Details

ECAI Conference 2020 Conference Paper

Learning Joint Job Embeddings Using a Job-Oriented Asymmetrical Pairing System

Timothé Bernard
Tanguy Moreau
Clément Viricel
Paul Mougel
Christophe Gravier
Frédérique Laforest

A main objective of Human Resources (HR) professionals is to suggest relevant job opportunities to people seeking change. This requires they identify key features in thousands of badly entitled, loosely-defined and often long job descriptions to compare job offers. However, an ever-increasing number of profiles and many similar-looking job opportunities combined with a quickly evolving job market make this task significantly harder and require advanced automatic tools. In this work, we tackle these problems by building a job embedding space called job2vec and a title predictor from a description. To do so, we propose the first Job-Oriented Asymmetrical Pairing System, JOAPS which on one hand uses a title encoder to create the job2vec and on the other hand uses a description encoder to predict a job title from a job description. The JOAPS architecture is based on two different sub-neural networks connected at their outputs using a novel weighted ranking pairwise loss, that we named Weighted On-the-fly Chunk Ranking (WOCR) loss, to back-propagate simultaneously to both sub-networks. First, we compare the job2vec built by JOAPS to a word2vec-based job2vec. Then we compare the JOAPS description encoder to a Sequence to Sequence baseline model. Our experiments show that JOAPS job2vec incorpore HR key features, whereas word2vec-based job2vec only combines features extracted from job title words. Thus JOAPS job2vec has more suitable properties to compare jobs, even with never-seen-before job titles. Futhermore, we show that JOAPS description encoder outperforms the Seq2Seq baseline method in title prediction.

Details

AAAI Conference 2019 Conference Paper

Near-Lossless Binarization of Word Embeddings

Julien Tissier
Christophe Gravier
Amaury Habrard

Word embeddings are commonly used as a starting point in many NLP models to achieve state-of-the-art performances. However, with a large vocabulary and many dimensions, these floating-point representations are expensive both in terms of memory and calculations which makes them unsuitable for use on low-resource devices. The method proposed in this paper transforms real-valued embeddings into binary embeddings while preserving semantic information, requiring only 128 or 256 bits for each vector. This leads to a small memory footprint and fast vector operations. The model is based on an autoencoder architecture, which also allows to reconstruct original vectors from the binary ones. Experimental results on semantic similarity, text classification and sentiment analysis tasks show that the binarization of word embeddings only leads to a loss of ∼2% in accuracy while vector size is reduced by 97%. Furthermore, a top-k benchmark demonstrates that using these binary vectors is 30 times faster than using real-valued vectors.

PDF Details

IJCAI Conference 2018 Conference Paper

Readitopics: Make Your Topic Models Readable via Labeling and Browsing

Julien Velcin
Antoine Gourru
Erwan Giry-Fouquet
Christophe Gravier
Mathieu Roche
Pascal Poncelet

Readitopics provides a new tool for browsing a textual corpus that showcases several recent work on topic labeling and topic coherence. We demonstrate the potential of these techniques to get a deeper understanding of the topics that structure different datasets. This tool is provided as a Web demo but it can be installed to experiment with your own dataset. It can be further extended to deal with more advanced topic modeling techniques.

PDF Details

IS Journal 2012 Journal Article

Adaptive System for Collaborative Online Laboratories

Christophe Gravier
Jacques Fayolle
Lardon
Martin J. O'Connor

Group activities are important aspects of the traditional laboratory experience. A framework using Semantic Web technologies supports collaborative strategies for online laboratories as well.

Details DOI