A Chinese Machine Reading Comprehension Dataset Automatic Generated Based on Knowledge Graph

A Chinese Machine Reading Comprehension Dataset Automatic Generated Based on Knowledge Graph Hanyu Zhao1, Yuan Sha 1, �, Jiahong Leng1, Xiang Pan2, Zhao Xue 1、Quanyue Ma 1 and Yangxiao Liang 1 1 Beijing Academy of Artificial Intelligence, Beijing, China 2 New York University, New York, America {hyzhao, yuansha, jhleng, xuezhao, maqy, yxliang}@baai.ac.cn xiangpan@nyu.edu Abstract Machine reading comprehension (MRC) is a typical natural language processing (NLP) task and has developed rapidly in the last few years. Various reading comprehension datasets have been built to support MRC studies. However, large-scale and high-quality datasets are rare due to the high complexity and huge workforce cost of making such a dataset. Besides, most reading comprehension datasets are in English, and Chinese datasets are insufficient. In this paper, we propose an automatic method for MRC dataset generation, and build the largest Chinese medical reading comprehension dataset presently named CMedRC. Our dataset contains 17k questions generated by our auto- matic method and some seed questions. We obtain the corresponding answers from a medical knowledge graph and manually check all of them. Finally, we test BiLSTM and BERT-based pre-trained language models (PLMs) on our dataset and propose a base- line for the following studies. Results show that the automatic MRC dataset generation method is considerable for future model improvements. 1 Introduction Medical care is closely related to people’s lives and helps people keep healthy in several different aspects, including disease prevention, medical examination, disease diagnosis, and treatment. Generally, various medical services are provided by those professionals who have specialized knowledge and rich experience. However, many countries now face severe medical personnel shortages, which means the demand for medical services dramatically exceeds the limit that professionals can supply. The rapidly developed artificial intelligence (AI) technology is a po- tential solution to the doctor shortage problem. AI medical experts are expected to offer various kinds of service for patients after learning enough knowledge from human experts and thus, re- duce doctors’ burden to a great extent. One of the potential uses of AI experts is to provide medical consultation for patients. The technical nature of such kinds of applications is the automated question answering technology. Recently, the automatic question answering system is a significant development direction of the intelligent medical industry. Retrieval-based question answering and knowledge base question answering (KBQA) are two primary forms of present medical QA systems. Retrieval type system selects the candidate in the existing collection of QA pairs with the highest similarity to the user’s input question. It returns the corresponding answer as the final response. As for KBQA, the system first extracts a topic entity in the user’s question, then searches it in the knowledge base and returns the most suitable neighbor of a topic entity as the final answer. However, the above two methods both have poor performances when answering new questions that have not been appeared in the dataset or knowledge base of the QA system. Thus, they are not adaptive to the medical industry, where new things appear almost all the …

Content truncated to 500 words