基于问句类型的问句相似度计算

Questions similarity computation based on question classification

目前,问句相似度的计算主要借鉴普通陈述句的相似度计算方法。由于普通陈述句的相似性更多反映的是语句间语义上的匹配符合程度,而衡量问句间的相似性则须同时考虑问句及其答案句之间的相似程度,为此,设计了一种新的问句相似度计算方法。该方法不仅利用问句之间的语义和语法特征考察问句之间的匹配程度,还利用问句的问题类型等信息来间接刻画答案句之间的特征形象,从而以获取问句的深层语义信息,以提高问句相似度计算的准确性。实验验证了该方法的有效性。

Now, questions similarity computation methods are mainly based on general declarative sentence similarity compu- tation methods, where only semantic similarity between two general declarative sentences is considered. While considering question similarity, both similarity between two questions and similarity between two answers should be considered, So this pa- per proposed a new question similarity computation method, which took into account not only semantic and syntactic characteri- stics between questions, but also information such as question type to indirectly get the characteristics of the answer, in order to extract deep semantic information of question, and finally to improve question similarity computation accuracy. Experimental results show that the method can achieve better accuracy.