基于常问问题集的中文问答系统研究  

Question answering system based on frequently asked questions

作　　者：秦兵[1] 刘挺[1] 王洋[1] 郑实福[1] 李生[1] 

机构地区：[1]哈尔滨工业大学计算机科学与技术学院,黑龙江哈尔滨150001

出　　处：《哈尔滨工业大学学报》2003年第10期1179-1182,共4页Journal of Harbin Institute of Technology

基　　金：国家自然科学基金资助项目(60203020);国家高技术发展计划资助项目(863-2001AA114041);哈尔滨工业大学校基金资助项目(HIT.2000 50).

摘　　要：首先根据用户的提问建立一个候选问题集,然后通过计算句子语义相似度,在候选问题集中找到相似的问句,并将答案返回给用户.该系统还能够自动地更新和维护FAQ库.实验表明,与基于关键词的句子相似度计算相比,基于语义的句子相似度计算提高了问题匹配的准确率.A candidate question set is built up according to the users' query. Semantic similarities of sentences are computed between the users' query and the candidate questions. The answer in correspondence with the most similar with query is thus returned to the user. This system can also automatically update and maintain FAQ. Experiments show that the new computing method gets a better performance than the keywords-based approach.

关 键 词：常问问题集 中文问答系统 候选问题集 句子相似度 数据库 数据结构 TFIDF方法 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]