Research On The Community-Driven Qa System Oriented Relatated Infor-Mation Recommendation Techniques

With the rapid growth of internet, there have been many network services for users to access information. Among of them, The community-driven question- answering (cQA) system is welcomed in that it provides a knowledge sharing platform based on interaction of users. It has accumulated a lot of question-answer resources, so users can search needed information without login. To get use of users'queries, a community recommendation system based on the semantic relationship between question-answer pairs is described in this paper. The system is presented to improve the performance in getting better information for users. It can recommend some people related to users and encourage them to join the cQA system for interaction. Our research focuses on the following aspects:Firstly, we study the semantic relationship between question-answer pairs. The multilayer neural networks consisting of RBM is introduced. We work on the optimization of structure and proper parameter settings for the neural networks to get dimension reduction of the question-answer features. Experimental results show the method goes well on the semantic relevance between questions and answers, and it helps to solve the ?lexical gap'problem in traditional keywords based information retrieval.Secondly, using cQA structure information for answer quality evaluation. Machine learning methods are used to get quality evaluation, while the community structure is described as feature for an answer. Experimental results show that, the community structure information is helpful to improve the performance of relationship between questions and answers.Thirdly, this paper researches user recommendation method based on similarity of question-answer pairs. We can get users'history questions from the community, and they can be represented as vectors by the multilayer neural network. The vectors are considered as a new kind of user information. In ordered to verify the effect of this information, a user-recommendation algorithm is presented, in which the input is a query (a question typed by user), and output is a list of users related to the input . The algorithm can help to attract new users registering for a cQA, because it can offer them interesting people in the cQA. In the experiment, we have volunteers to evaluate the recommendation method, the result shows that it gets a good performance.