A Hybrid Multi-answer Summarization Model for the Biomedical Question-Answering System

Abstract

In natural language processing problems, text summarization is a difficult problem and always attracts attention from the research community, especially working on biomedical text data which lacks supporting tools and techniques. In this scientific research report, we propose a multi-document summarization model for the responses in the biomedical question and answer system. Our model includes components which is a combination of many advanced techniques as well as some improved methods proposed by authors. We present research methods applied to two main approaches: an extractive summarization architecture based on multi scores and state-of-the-art techniques, presenting our novel prosper-thy-neighbor strategies to improve performance; EAHS model (Extractive-Abstractive hybrid model) based on a denoising auto-encoder for pre-training sequence-to-sequence models (BART). In which we propose a question-driven filtering phase to optimize the selection of the most useful information. Our propose model has achieved positive results with the best ROUGE-1/ROUGE-L scores being the runner-up by ROUGE-2 $F1$ score by extractive summarization results (over 24 participated teams in MEDIQA2021).