基于多视角匹配的中文问答对自动生成框架  

AUTOMATIC GENERATION FRAMEWORK OF CHINESE QUESTION AND ANSWER PAIR BASED ON MULTI-VIEW MATCHING

作　　者：尹文峰 黄莉[1,2,3] 顾进广 Yin Wenfeng;Huang Li;Gu Jinguang(School of Computer Science and Technology,Wuhan University of Science and Technology,Wuhan 430065,Hubei,China;Big Data Science and Engineering Research Institute,Wuhan University of Science and Technology,Wuhan 430065,Hubei,China;Hubei Province Key Laboratory of Intelligent Information Processing and Real-Time Industrial,Wuhan 430065,Hubei,China)

机构地区：[1]武汉科技大学计算机科学与技术学院,湖北武汉430065 [2]武汉科技大学大数据科学与工程研究院,湖北武汉430065 [3]湖北省智能信息处理与实时工业系统重点实验室,湖北武汉430065

出　　处：《计算机应用与软件》2024年第3期163-168,共6页Computer Applications and Software

基　　金：国家自然科学基金项目(61673304);国家社科基金重大计划项目(11&ZD189);湖北省自然科学基金项目(2018CFB194)。

摘　　要：针对目前问答对生成方法中问题与答案不完全匹配的问题,提出一种基于神经网络自动从中文生成问答对的方法。使用命名实体识别和规则的方法从文本中抽取关键词,确定问题的主题;使用多视角匹配的神经网络模型从文本中生成问题,避免对手工模板强依赖;使用阅读理解模型根据问题生成置信度更高的答案。实验结果分析表明,生成问题的质量高于基于模板的方法,并且能够过滤80%的不匹配问答对。Aimed at the problem that the question and the answer in the current question and answer generation method do not completely match,a method for automatically generating question and answer pairs from Chinese text based on neural network is proposed.We used the method of named entity recognition and rules to extract keywords from the text to determine the topic of the problem,and used the neural network model of multi-view matching to generate the problem from the text,avoiding strong dependence on manual templates.We used the reading comprehension model according to the problem to generate answers with higher confidence.The results show that the quality of the generated questions is higher than that of the template-based method,and 80%of unmatched question and answer pairs can be filtered.

关 键 词：中文 问题生成 神经网络 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]