Research On Sentences Similarity Computation Based On Multi-information Fusion

Sentence similarity computation is an essential research and directly decides the development of certain other related fields.For example:in the fields of Automatic Question-Answer,Machine Translation,Smart Search,sentence similarity computation is one of the most important technologies.In this paper,similarity computation we have studied is focusing on three levels:sememe,word and sentence,which focuses on the sentence similarity computation.The method of sentences similarity computation based on multi-information fusion is proposed basing on the current inadequacy of the sentence similarity computation and designs a experiment to prove the validity of the method;Finally,gives the application of sentence similarity computation in FAQ combining with the Automatic Question-Answer.The paper has the following main aspects:1.Researchs on the word similarity and the method of it,by using rich semantic information provided by the "HowNet" to compute the sememe similarity,and further improves the method of word similarity computation based on the "HowNet".2.Proposes the method of sentence similarity computation based on multi-information fusion.The method is from different angles considering the information of keywords,semantic,syntactic structure,extracted from the sentence length,the same number of keywords,the distance of keywords,the sequence of keywords and other informations,using a weighted sum to determine sentence similarity.3.Uses Genetic Algorithms to solve Keywords information,semantic information and syntactic structure information between the informations the best combination of weight to avoid the adoption of the right to determine the value of the experience of the unreliability and subjectivity,and optimize the portfolio weights. 4.Designs a Bank Area Automatic Question-Answer(BAQS) model and gives the application of sentence similarity computation,sentence similarity computation for bank area Automatic Question-Answer matching FAQ questions database shows the realization process of the sentence similarity computation in the FAQ database.