The Study On Routing Questions In Community Question Answering

The rapid development of internet technology brings great convenience for people's daily life. A large amount of information makes people lost in the sea of information. It will lead that users spend too much time to find the information they need. With the development of Web2.0, search engine has not solved the professional issues exactly and users cannot receive good interactive experience. In recent years a large number of community question answering systems come into being, make up drawbacks of search engine and meet different area needs. In community question answering users can provide themselves questions and then wait others to answer their questions which will take several days and the result may be incorrect?attack or offensive answers. Users can search related questions with their questions to receive answers. Historical archives of community question answering contain limited answers set and users have to deal with the word-match constraint between her formulated question and archived questions. Therefore, a kind method of routing questions to experts is necessary.First of all, some methods of routing questions use classical information retrieval approaches that retrieve good results if sufficient word overlap exists. However there are few overlap words between most of new questions and user profile, therefore the result is not satisfactory. In this paper we propose one method of routing questions to the best answerers based on words distributed representations. In community question answering system tags can represent user topic, we adopt tags to extract topic words and train large amounts of data to obtain words distributed representations. We leverage topic words distributed representation to indicate similarity between user profile and new question. Thereby we rout questions to the experts. We compare our method with other methods, and the experimental results are better than other methods.Secondly, the supervised convolutional neural network model has a good effect in sentence classification. We classify new questions based on that characteristic and predict the best answerer. We build profile for each user and the profile is consist of the questions in which the user be chosen the best answer. In classification model classifies correspond to candidate answerer. Therefore we can obtain probabilities of every user being the best answerer for new questions. Based on the probabilities we rout questions to the experts. The experimental results show our methods are improved significantly than other methods.