Recent studies introduced effective compression techniques for Large Language Models (LLMs) via post-training quantization or low-bit weight representation. Although quantized weights offer storage efficiency and allow for faster inference, existing works have indicated that quantization might compromise performance and exacerbate biases in LLMs.This study investigates the confidence and calibration of quantized models, considering factors such as language model type and scale as contributors to quantization loss.Firstly, we reveal that quantization with GPTQ to 4-bit results in a decrease in confidence regarding true labels, with varying impacts observed among different language models. Secondly, we observe fluctuations in the impact on confidence across different scales. Finally, we propose an explanation for quantization loss based on confidence levels, indicating that quantization disproportionately affects samples where the full model exhibited low confidence levels in the first place.We make our code and quantized models publicly available.

Learning authors representations from their textual productions is now widely used to solve multiple downstream tasks, such as classification, link prediction or user recommendation. Author embedding methods are often built on top of either Doc2Vec (Mikolov et al. 2014) or the Transformer architecture (Devlin et al. 2019). Evaluating the quality of these embeddings and what they capture is a difficult task. Most articles use either classification accuracy or authorship attribution, which does not clearly measure the quality of the representation space, if it really captures what it has been built for. In this paper, we propose a novel evaluation framework of author embedding methods based on the writing style. It allows to quantify if the embedding space effectively captures a set of stylistic features, chosen to be the best proxy of an author writing style. This approach gives less importance to the topics conveyed by the documents. It turns out that recent models are mostly driven by the inner semantic of authors’ production. They are outperformed by simple baselines, based on state-of-the-art pretrained sentence embedding models, on several linguistic axes. These baselines can grasp complex linguistic phenomena and writing style more efficiently, paving the way for designing new style-driven author embedding models.

Slow emerging topic detection is a task between event detection, where we aggregate behaviors of different words on short period of time, and language evolution, where we monitor their long term evolution. In this work, we tackle the problem of early detection of slowly emerging new topics. To this end, we gather evidence of weak signals at the word level. We propose to monitor the behavior of words representation in an embedding space and use one of its geometrical properties to characterize the emergence of topics. As evaluation is typically hard for this kind of task, we present a framework for quantitative evaluation and show positive results that outperform state-of-the-art methods. Our method is evaluated on two public datasets of press and scientific articles.

Cet article s’intègre dans un projet collaboratif qui vise à réaliser une analyse longitudinale de la production universitaire en Géographie. En particulier, nous présentons les premiers résultats de l’application d’une méthode de détection automatique de métaphores basée sur les modèles de thématiques latentes. Une analyse détaillée permet de mieux comprendre l’impact de certains choix et de réfléchir aux pistes de recherche que nous serons amenés à explorer pour améliorer ces résultats.

Ce travail concerne l’intégration à une plateforme de veille sur internet d’outils permettant l’analyse des opinions émises par les internautes à propos d’une entité, ainsi que la manière dont elles évoluent dans le temps. Les entités considérées peuvent être des personnes, des entreprises, des marques, etc. Les outils implémentés sont le produit d’une collaboration impliquant plusieurs partenaires industriels et académiques dans le cadre du projet ANR ImagiWeb.

The objective of this paper is to describe the design of a dataset that deals with the image (i.e., representation, web reputation) of various entities populating the Internet: politicians, celebrities, companies, brands etc. Our main contribution is to build and provide an original annotated French dataset. This dataset consists of 11527 manually annotated tweets expressing the opinion on specific facets (e.g., ethic, communication, economic project) describing two French policitians over time. We believe that other researchers might benefit from this experience, since designing and implementing such a dataset has proven quite an interesting challenge. This design comprises different processes such as data selection, formal definition and instantiation of an image. We have set up a full open-source annotation platform. In addition to the dataset design, we present the first results that we obtained by applying clustering methods to the annotated dataset in order to extract the entity images.