Application And Realization Of Sentence Similarity Matching In Automatic Question Answering System

With the rapid development of artificial intelligence technology,application systems such as robots,intelligent interactive terminals,and smart call centers have been gradually presented in the real world.Among them,intelligent voice interaction,that is,automatic question and answer has become an important module for various types of smart system applications.The FAQ-oriented question-answering system,which organizes the frequently asked questions and the corresponding answers,provides the user with a more convenient solution to the problem.In the automatic question answering system,sentence similarity matching is the key technology.The thesis studies sentence similarity calculation from the perspective of syntactic structure and co-occurrence of words,and proposes a sentence similarity calculation method based on word vectors for multi-feature fusion.The thesis mainly does the following work in the process of sentence similarity matching: The word similarity based on Word2 Vec word vector is used to calculate the similarity of sentences.The word vector can more accurately reflect the semantic relationship of words and provides a good foundation for sentence semantic calculation.The use of sentence-dependent structure analysis tools to determine the center of the word and sentence components,access to dependency pairs with the relationship.In order to highlight the importance of different dependency pairs in sentences,the dissertation sets the importance degree to dependency pairs and effectively avoids the influence of Chinese complex sentences on sentence similarity calculation.On the basis of co-occurrence of words,word vectors are used to calculate the similarity of sentences.Finally,the two algorithms are combined and given different weights.In order to verify the validity of this method,the sentence similarity calculation method is applied to the automatic question answering system based on the application of government lobby.Through the comparative analysis of the experiment,this thesis verifies the proposed sentence similarity calculation method of multi-feature fusion.Through actual system verification tests,it is shown that the accuracy and recall rate of the proposed method are better when calculating the similarity of sentences.