||Similarity Calculation and Proposed Benefit-harm Value for Grammar Learning Chat Robot
102015-03-24 , 法政大学大学院情報科学研究科
Traditional chat robot systems mainly use semantic analysis and similarity calculation to achieve the purpose of communicating with users. Semantic analysis is applied to analyze grammar structure of input and to return a grammatically correct output. It needs an enormous rule base of grammar built by language experts. Similarity calculation compares how similar input and sentences restored in database are and finds keywords for output. However there are two open problems in this kind of systems. First, output returned by traditional chat robot system cannot contain all key information existing in database. Second, the rule base of grammar is permanent thus lacking an ability to learn new grammars if necessary. This paper proposes a novel chat robot system whose output contains all possible key information in database and whose rule base of grammar can grow through communication with users. We take two experiments to compare the properity of traditional chat robot system and proposed chat robot system. First experiment tests how many necessary keywords the outputs of traditional chat robot system and proposed one can contain respectively. Second experiment asks 10 volunteers to give score to both systems in order to evaluate their rule base of grammar.