依赖关系三元组匹配在问题回答中的应用

Dependency Relation Triples Matching for Question Answering

描述了一种新的计算问题与支持答案句相似度的方法,即基于依赖关系三元组匹配的方法.该方法引入了问题中的疑问性和非疑问性部分的信息,采用了启发式规则扩展问题的依赖关系三元组,从而匹配变形的答案句.同时把问题与支持答案句的相似度作为新的特征,应用于开放领域的问题回答(Question answering,QA)任务中的答案排序.实验结果表明,引入新特征的答案排序方法与通常的基于密度的方法相比,在相对精度指标上提高了8.2%,在平均排序倒数(Mean reciprocal rank,MRR)评价上提高了8%.

This paper presents a new method to compute the similarity between question and answer sentences, namely dependency relation triples matching. This method considers the information of questionls interrogative part and noninterrogative part, and heuristic rules are used to expand questionls relation triples to match metamorphosing answer sentences. Then, this similarity score is used as a new feature for answer ranking in open domain question answering （QA） track. The experiments show the new answer ranking method outperforms the common density-based approach by up to 8.2 % in relative precision and 8 % in mean reciprocal rank （MRR） evaluation.