语义相似计算驱动领域自动问答

Domain-Specific QA Driven by Computation of Semantic Similarity

  • 摘要: 研究实体相似性的认知心理特征和受限领域自动问答(QA)系统的形式特点.基于结构对齐和几何相似模型,把词语概念描述分解为属性部分和语义角色部分,各部分结构分别对齐后,计算对齐义原的关系距离,加权组合计算词语的相似度.该方法也适用于解析成语义向量表示的疑问句的相似度计算.融合通用本体、领域本体和领域知识文本,构造了支持语义计算求解某一银行QA问题的知识库.实验表明,该方法可以提高领域QA系统的用户满意度.

     

    Abstract: Characteristics of cognitive psychology about entity similarity and formal features of Domain-specific question-answering systems are presented.Based on the structural-alignment and geometric similarity model,a new approach to Chinese word similarity computation is proposed.By separating the word concept description into two parts of attribute space and case space,which respectively process the structural alignment and calculate the distance of sememes aligned,the word concept similarity is made of the two parts distance with different weight.The method is applicable to compute the similarity of two sentences represented in semantic vector(quasi-case frame).The world ontology,domain ontology and domain knowledge text are fused to make up the knowledge base for semantic computation,and thus to drive a specific bank QA system.Experimental results showed that the method is capable of improving user satisfaction of QA systems.

     

/

返回文章
返回
Baidu
map