Lynda Tamine

Jose G. Moreno, Jesus Lovon, M'Rick Robin-Charlet et al.

Fine-tuning of Large Language Models (LLMs) has become the default practice for improving model performance on a given task. However, performance improvement comes at the cost of training on vast amounts of annotated data which could be sensitive leading to significant data privacy concerns. In particular, the healthcare domain is one of the most sensitive domains exposed to data privacy issues. In this paper, we present PatientDx, a framework of model merging that allows the design of effective LLMs for health-predictive tasks without requiring fine-tuning nor adaptation on patient data. Our proposal is based on recently proposed techniques known as merging of LLMs and aims to optimize a building block merging strategy. PatientDx uses a pivotal model adapted to numerical reasoning and tunes hyperparameters on examples based on a performance metric but without training of the LLM on these data. Experiments using the mortality tasks of the MIMIC-IV dataset show improvements up to 7% in terms of AUROC when compared to initial models. Additionally, we confirm that when compared to fine-tuned models, our proposal is less prone to data leak problems without hurting performance. Finally, we qualitatively show the capabilities of our proposal through a case study. Our best model is publicly available at https://huggingface.co/ Jgmorenof/mistral\_merged\_0\_4.

Jesus Lovon-Melgarejo, Jose G. Moreno, Christine Damase-Michel et al.

Electronic Health Records (EHRs) provide crucial information for clinical decision-making. However, their high-dimensionality, heterogeneity, and sparsity make clinical prediction challenging. Large Language Models (LLMs) allowed progress towards addressing this challenge by leveraging parametric medical knowledge to enhance EHR data for clinical prediction tasks. Despite the significant achievements made so far, most of the existing approaches are fundamentally task-agnostic in the sense that they deploy LLMs as EHR encoders or EHR completion modules without fully integrating signals from the prediction tasks. This naturally hinders task performance accuracy. In this work, we propose Rewrite-To-Predict (ReToP), an LLM-based framework that addresses this limitation through an end-to-end training of an EHR rewriter and a clinical predictor. To cope with the lack of EHR rewrite training data, we generate synthetic pseudo-labels using clinical-driven feature selection strategies to create diverse patient rewrites for fine-tuning the EHR rewriter. ReToP aligns the rewriter with prediction objectives using a novel Classifier Supervised Contribution (CSC) score that enables the EHR rewriter to generate clinically relevant rewrites that directly enhance prediction. Our ReToP framework surpasses strong baseline models across three clinical tasks on MIMIC-IV. Moreover, the analysis of ReToP shows its generalizability to unseen datasets and tasks with minimal fine-tuning while preserving faithful rewrites and emphasizing task-relevant predictive features.

Jesus Lovon, Thouria Ben-Haddi, Jules Di Scala et al.

The lack of standardized evaluation benchmarks in the medical domain for text inputs can be a barrier to widely adopting and leveraging the potential of natural language models for health-related downstream tasks. This paper revisited an openly available MIMIC-IV benchmark for electronic health records (EHRs) to address this issue. First, we integrate the MIMIC-IV data within the Hugging Face datasets library to allow an easy share and use of this collection. Second, we investigate the application of templates to convert EHR tabular data to text. Experiments using fine-tuned and zero-shot LLMs on the mortality of patients task show that fine-tuned text-based models are competitive against robust tabular classifiers. In contrast, zero-shot LLMs struggle to leverage EHR representations. This study underlines the potential of text-based approaches in the medical field and highlights areas for further improvement.

Jesús Lovón-Melgarejo, Jose G. Moreno, Romaric Besançon et al.

Since Pretrained Language Models (PLMs) are the cornerstone of the most recent Information Retrieval (IR) models, the way they encode semantic knowledge is particularly important. However, little attention has been given to studying the PLMs' capability to capture hierarchical semantic knowledge. Traditionally, evaluating such knowledge encoded in PLMs relies on their performance on a task-dependent evaluation approach based on proxy tasks, such as hypernymy detection. Unfortunately, this approach potentially ignores other implicit and complex taxonomic relations. In this work, we propose a task-agnostic evaluation method able to evaluate to what extent PLMs can capture complex taxonomy relations, such as ancestors and siblings. The evaluation is based on intrinsic properties that capture the hierarchical nature of taxonomies. Our experimental evaluation shows that the lexico-semantic knowledge implicitly encoded in PLMs does not always capture hierarchical relations. We further demonstrate that the proposed properties can be injected into PLMs to improve their understanding of hierarchy. Through evaluations on taxonomy reconstruction, hypernym discovery and reading comprehension tasks, we show that the knowledge about hierarchy is moderately but not systematically transferable across tasks.

Jesus Lovon, Martin Mouysset, Jo Oleiwan et al.

Electronic Health Record (EHR) tables pose unique challenges among which is the presence of hidden contextual dependencies between medical features with a high level of data dimensionality and sparsity. This study presents the first investigation into the abilities of LLMs to comprehend EHRs for patient data extraction and retrieval. We conduct extensive experiments using the MIMICSQL dataset to explore the impact of the prompt structure, instruction, context, and demonstration, of two backbone LLMs, Llama2 and Meditron, based on task performance. Through quantitative and qualitative analyses, our findings show that optimal feature selection and serialization methods can enhance task performance by up to 26.79% compared to naive approaches. Similarly, in-context learning setups with relevant example selection improve data extraction performance by 5.95%. Based on our study findings, we propose guidelines that we believe would help the design of LLM-based models to support health search.

Lucas Albarede, Jose Moreno, Lynda Tamine et al.

Despite their impressive performances, Large Language Models (LLMs) remain prone to hallucination, which critically undermines their trustworthiness. While most of the previous work focused on tackling answer and attribution correctness, a recent line of work investigated faithfulness, with a focus on leveraging internal model signals to reflect a model's actual decision-making process while generating the answer. Nevertheless, these methods induce additional latency and have shown limitations in directly aligning token generation with attribution generation. In this paper, we introduce LoDIT, a method that jointly generates and faithfully attributes answers in RAG by leveraging specific token logits during generation. It consists of two steps: (1) marking the documents with specific token identifiers and then leveraging the logits of these tokens to estimate the contribution of each document to the answer during generation, and (2) aggregating these contributions into document attributions. Experiments on a trustworthiness-focused attributed text-generation benchmark, Trust-Align, show that LoDIT significantly outperforms state-of-the-art models on several metrics. Finally, an in-depth analysis of LoDIT shows both its efficiency in terms of latency and its robustness in different settings.

Hanane Djeddal, Thomas Gerald, Laure Soulier et al.

In this work, our aim is to provide a structured answer in natural language to a complex information need. Particularly, we envision using generative models from the perspective of data-to-text generation. We propose the use of a content selection and planning pipeline which aims at structuring the answer by generating intermediate plans. The experimental evaluation is performed using the TREC Complex Answer Retrieval (CAR) dataset. We evaluate both the generated answer and its corresponding structure and show the effectiveness of planning-based models in comparison to a text-to-text model.

Jesus Lovon-Melgarejo, Laure Soulier, Karen Pinel-Sauvagnat et al.

Several deep neural ranking models have been proposed in the recent IR literature. While their transferability to one target domain held by a dataset has been widely addressed using traditional domain adaptation strategies, the question of their cross-domain transferability is still under-studied. We study here in what extent neural ranking models catastrophically forget old knowledge acquired from previously observed domains after acquiring new knowledge, leading to performance decrease on those domains. Our experiments show that the effectiveness of neuralIR ranking models is achieved at the cost of catastrophic forgetting and that a lifelong learning strategy using a cross-domain regularizer success-fully mitigates the problem. Using an explanatory approach built on a regression model, we also show the effect of domain characteristics on the rise of catastrophic forgetting. We believe that the obtained results can be useful for both theoretical and practical future work in neural IR.

Gia-Hung Nguyen, Laure Soulier, Lynda Tamine et al.

The state-of-the-art solutions to the vocabulary mismatch in information retrieval (IR) mainly aim at leveraging either the relational semantics provided by external resources or the distributional semantics, recently investigated by deep neural approaches. Guided by the intuition that the relational semantics might improve the effectiveness of deep neural approaches, we propose the Deep Semantic Resource Inference Model (DSRIM) that relies on: 1) a representation of raw-data that models the relational semantics of text by jointly considering objects and relations expressed in a knowledge resource, and 2) an end-to-end neural architecture that learns the query-document relevance by leveraging the distributional and relational semantics of documents and queries. The experimental evaluation carried out on two TREC datasets from TREC Terabyte and TREC CDS tracks relying respectively on WordNet and MeSH resources, indicates that our model outperforms state-of-the-art semantic and deep neural IR models.

Gia-Hung Nguyen, Lynda Tamine, Laure Soulier et al.

This paper tackles the problem of the semantic gap between a document and a query within an ad-hoc information retrieval task. In this context, knowledge bases (KBs) have already been acknowledged as valuable means since they allow the representation of explicit relations between entities. However, they do not necessarily represent implicit relations that could be hidden in a corpora. This latter issue is tackled by recent works dealing with deep representation learn ing of texts. With this in mind, we argue that embedding KBs within deep neural architectures supporting documentquery matching would give rise to fine-grained latent representations of both words and their semantic relations. In this paper, we review the main approaches of neural-based document ranking as well as those approaches for latent representation of entities and relations via KBs. We then propose some avenues to incorporate KBs in deep neural approaches for document ranking. More particularly, this paper advocates that KBs can be used either to support enhanced latent representations of queries and documents based on both distributional and relational semantics or to serve as a semantic translator between their latent distributional representations.

Bilel Moulahi, Lynda Tamine, Sadok Ben Yahia

In light of the tremendous amount of data produced by social media, a large body of research have revisited the relevance estimation of the users' generated content. Most of the studies have stressed the multidimensional nature of relevance and proved the effectiveness of combining the different criteria that it embodies. Traditional relevance estimates combination methods are often based on linear combination schemes. However, despite being effective, those aggregation mechanisms are not effective in real-life applications since they heavily rely on the non-realistic independence property of the relevance dimensions. In this paper, we propose to tackle this issue through the design of a novel fuzzy-based document ranking model. We also propose an automated methodology to capture the importance of relevance dimensions, as well as information about their interaction. This model, based on the Choquet Integral, allows to optimize the aggregated documents relevance scores using any target information retrieval relevance metric. Experiments within the TREC Microblog task and a social personalized information retrieval task highlighted that our model significantly outperforms a wide range of state-of-the-art aggregation operators, as well as a representative learning to rank methods.