Research On Question Retrieval Method In Community Question Answering System

One notable feature of Web 2.0 is that the provider of Internet content has changed from the former administrator to the entire Internet user.There are countless User Generated Content(UGC)submissions on the Internet every day.This enriches the network content on the one hand,and on the other it also floods the network with a lot of noise,giving users who want to acquire information knowledge as soon as possible from internet cause some difficulties.The emergence of a community-based question-answering system based on user-generated content solves this problem.It relies on the participation of all community users to solve the same issues raised by community users.This type of community user collaboration has become an important source of knowledge for the Internet.After a period of development,the community question answering system has accumulated a large number of users have solved problem,how to utilize the answers to the questions in the question and answer community to answer the user's newly submitted questions,become a major topic of question answering system research,which is called question retrieval.Question retrieval on the one hand can help users quickly obtain the answers to the questions.On the other hand,it can also alleviate the problem of redundant questions that are submitted to the problem system.The second chapter of this paper proposes a question retrieval model based on the translation model,which solves the problem that the question retrieval performance of the translation model is vulnerable to translation noise when the quality of the background corpus is not high.Specifically,a translation model is a statistical retrieval model that trains the translation probabilities of terms through large-scale background corpus.In an ideal situation,the trained translation model can recognize the translation probabilities between words,and then determine the similarity between words and words.However,due to the lack of such a large-scale corpus,the retrieval efficiency of the translation model is not very satisfactory.On the other hand,the use of the community attributes of the community question answering system to collect parallel questions with a high degree of similarity is used as the background corpus.This is mainly due to the similar issues of classification tags and user tags of questions,and on the other hand,it is easy to use the knowledge of lexical semantics provided by HNC.The translation probability of the deviation.Based on the above work,a question retrieval model which integrates knowledge of HNC lexical semantics is proposed.The experimental results show that the retrieval effect of the model proposed in this paper is better than some models with better effect.The third chapter of this paper proposes a question retrieval model that combines three levels of similarities: pragmatics,grammar and semantics.It calculates the similarity relationship between query questions and candidate questions from the whole sentence,and has a similar semantic level.The degree calculation uses the method proposed in Chapter 3.Specifically,in the pragmatic level similarity calculation,this paper uses the HIT question classification system to identify the pragmatic types of questions and then calculate the pragmatic similarity of sentences.In the calculation of the grammatical similarity,the syntactic similarity relationship between the sentences is obtained by using the sentence class expression in the results of the HNC sentence class analysis and by comparing the structure of the sentence class expressions in the question sentence,and then using the question sentence class expressions.The semantic block consists of computing the semantic similarity between questions.After synthesizing the results of the above three similarity calculations,a new question retrieval model is proposed.The experiment on the real dataset verifies that the retrieval effect of the model is better than the previous retrieval model.