Medical Question Summarization with Entity-driven Contrastive Learning

TL;DR

This work addresses MQS by proposing Entity-Driven Contrastive Learning (ECL), a framework that uses medical entities as question focuses to generate hard negative samples and guide a MoCo-based contrastive objective alongside a BART-based summarizer. By replacing focuses with unrelated entities and leveraging simple negatives, ECL sharpens semantic representations and highlights essential medical terms, achieving state-of-the-art ROUGE scores across four datasets while mitigating data leakage through dataset reorganization. The approach also introduces targeted evaluations of focus identification and medical-entity consistency, demonstrating substantial gains over prior QFCL and BART-based baselines. Overall, ECL provides a principled, entity-centric method to improve MQS accuracy and reliability, with practical implications for health-question answering systems and benchmarking in MQS research.

Abstract

By summarizing longer consumer health questions into shorter and essential ones, medical question-answering systems can more accurately understand consumer intentions and retrieve suitable answers. However, medical question summarization is very challenging due to obvious distinctions in health trouble descriptions from patients and doctors. Although deep learning has been applied to successfully address the medical question summarization (MQS) task, two challenges remain: how to correctly capture question focus to model its semantic intention, and how to obtain reliable datasets to fairly evaluate performance. To address these challenges, this paper proposes a novel medical question summarization framework based on entity-driven contrastive learning (ECL). ECL employs medical entities present in frequently asked questions (FAQs) as focuses and devises an effective mechanism to generate hard negative samples. This approach compels models to focus on essential information and consequently generate more accurate question summaries. Furthermore, we have discovered that some MQS datasets, such as the iCliniq dataset with a 33% duplicate rate, have significant data leakage issues. To ensure an impartial evaluation of the related methods, this paper carefully examines leaked samples to reorganize more reasonable datasets. Extensive experiments demonstrate that our ECL method outperforms the existing methods and achieves new state-of-the-art performance, i.e., 52.85, 43.16, 41.31, 43.52 in terms of ROUGE-1 metric on MeQSum, CHQ-Summ, iCliniq, HealthCareMagic dataset, respectively. The code and datasets are available at https://github.com/yrbobo/MQS-ECL.

Figures (4)

• Figure 1: Framework of the proposed medical question summarization with entity-driven contrastive learning (ECL). It involves basic summarization module (BART), contrastive learning module (MoCo), simple negative sample module and hard negative sample module.
• Figure 2: Generation of hard and simple negative samples.
• Figure 3: Comparison of Focus Identification Rate (FIR). The x-axis represents the dataset, and the y-axis represents the FIR
• Figure 4: The procedure of similarity optimization during ECL training. The x-axis represents epoches of model training, and the y-axis represents similarities between CHQ and hard/simple negative samples. The dotted line in green indicates that the model performs best on the validation set at the corresponding epoch.