Medical Question Understanding and Answering with Knowledge Grounding and Semantic Self-Supervision

Proceedings of the 29th International Conference on Computational Linguistics, pages 2734–2747 October 12–17, 2022. 2734 Medical Question Understanding and Answering with Knowledge Grounding and Semantic Self-Supervision Khalil Mrini1, Harpreet Singh1, Franck Dernoncourt2, Seunghyun Yoon2, Trung Bui2, Walter Chang2, Emilia Farcas1, and Ndapa Nakashole1 1University of California, San Diego, La Jolla, CA 92093 {khalil, h1singh, efarcas, nnakashole}@ucsd.edu 2Adobe Research, San Jose, CA 95110 {franck.dernoncourt, syoon, bui, wachang}@adobe.com Abstract Current medical question answering systems have difficulty processing long, detailed and informally worded questions submitted by patients, called Consumer Health Questions (CHQs). To address this issue, we introduce a medical question understanding and answering system with knowledge grounding and seman- tic self-supervision. Our system is a pipeline that first summarizes a long, medical, user- written question, using a supervised summa- rization loss. Then, our system performs a two-step retrieval to return answers. The sys- tem first matches the summarized user question with an FAQ from a trusted medical knowl- edge base, and then retrieves a fixed number of relevant sentences from the corresponding answer document. In the absence of labels for question matching or answer relevance, we de- sign 3 novel, self-supervised and semantically- guided losses. We evaluate our model against two strong retrieval-based question answering baselines. Evaluators ask their own questions and rate the answers retrieved by our baselines and own system according to their relevance. They find that our system retrieves more rele- vant answers, while achieving speeds 20 times faster. Our self-supervised losses also help the summarizer achieve higher scores in ROUGE, as well as in human evaluation metrics. We re- lease our code to encourage further research.1 1 Introduction Motivation. Users of medical question answering systems often write long questions, called Con- sumer Health Questions (CHQs). Several aspects of CHQs hinder the capacity of current question an- swering (QA) systems to process them: long medi- cal questions may contain peripheral information like patient history (Roberts and Demner-Fushman, 2016) that are not necessary to retrieve relevant answers. Consumer health questions may also use 1Link: https://github.com/KhalilMrini/ Medical-Question-Answering Consumer Health Question Abstractive Summarizer Model Generated Summary Knowledge base of FAQs and Answers Answer Document Encoder Selected Answer Sentences 1. Reference FAQ Summarization Loss 2. Matching FAQ Similarity Loss 3. Answer Selection Loss 4. Answer Similarity Loss Figure 1: Overview of our proposed Consumer Health Question Understanding and Answering model. The input is a user question, called Consumer Heath Ques- tion (CHQ). The goal is to match the CHQ to relevant answer sentences associated with a Frequently Asked Question (FAQ) from a medical knowledge base. a distinct vocabulary from the one used by medi- cal providers to describe the same health concepts (Ben Abacha and Demner-Fushman, 2019a). A growing number of approaches attempt to en- hance the processing of consumer health questions – or medical question understanding. These ap- proaches include query relaxation (Ben Abacha and Zweigenbaum, 2015; Lei et al., 2020), question en- tailment (Ben Abacha and Demner-Fushman, 2016, 2019b; Agrawal et al., 2019), question summariza- tion (Ben Abacha and Demner-Fushman, 2019a), …

Content truncated to 500 words