University intelligent question answering system based on multi-layer strategy

针对高校校园这一应用领域，设计并实现了一种基于多层策略的校园问答系统。提出了校园知识的3种类型，即服务型知识、常见问题型知识及文本检索型知识。针对不同类型的知识，建立了特定服务查询、常见问题查询和自由文本检索多级策略的问答系统模型，逐级匹配查询问句的答案：特定服务查询通过模式匹配将问句转换为服务接口；常见问题查询通过特征关键词提取、问题分类和问句相似度计算算法将问句与常见问题库中的问题‐答案对匹配；自由文本检索对全文检索引擎solr返回的候选文档进行答案抽取，获取答案段落。性能测试表明：对80％以上的查询问句，若问题相关知识存在于系统中，则系统都能给出满意的答案。In the view of the university application domain ,a intelligent question‐answering system based on multi‐layer strategy was designed and implemented .Three types of university knowledge were proposed ,which were service knowledge ,frequently asking‐problems knowledge and full‐text retrieval knowledge .For different types of knowledge ,three layers strategy module ,which named specific service query layer ,frequently asking‐problems query layer and full‐text retrieval layer ,was established to find the answ er to the question step by step .T he specific service query layer converted the question as a application programming interface by pattern matching ;the frequently asking‐prob‐lems query layer matched question to question‐answering couple by feature keywords extraction ,ques‐tion classification and sentence similarity calculation algorithm ;the specific service query layer used the full text retrieval engine solr to get the candidate documents ,and then answer paragraph from candidate documents was obtained by answer extraction .The performance test indicate that for more than 80% of query ,the proposed system can give a satisfactory answer if related knowledge exists in the system .国家高技术研究发展计划资助项目(2015AA01A202)．中文核心期刊要目总览(PKU)中国科技核心期刊(ISTIC)中国科学引文数据库(CSCD)11117-1224