Research On Problem Classification And Recommendation Mechanism In Community Q & A System

Community question answering (Yahoo! Answers, Baidu, Sina love to ask, etc.) is one of a new information-sharing mode. In the community, the user is not only a questioner and also a answerer, the problem is the link between users. Community Q & A system set-level categories, allows users to automatically or manually when the category of questions are marked, and the user select the category you’re interested in to answer questions. However as the category of questions expansion and refinement, rely on artificial markers is both time-consuming and inefficient, so most questioners unmarked the category of questions, if only use a small part of questions which artificial marking categories, through to question classification automatically by the traditional classification method can reduce the accuracy of classification. Moreover, as the number of users and amount of questions and answers the growing surge, the question raised by one second before a user is likely to issue a second later be presented to other users to drown, resulting in a significant reduction in the efficiency question is answered. Therefore, this research around key issues which questions classified and personalized recommendation mechanism, accomplish the following main research work:(1) Using a graph-based thinking, we proposed the use of the label propagation algorithm to classify questions. This method firstly use map strategies to establish a question classification. In this graph model, each node represented have tag question samples or no tag question samples, and the edge which between node and node represents the similarity of question samples. Then, question classification is converted into the node’s label transfer to the other nodes through transmission probability, propagation of circulating until convergence in marked probability, question classification is completed. Use "Badu Knows" crawl on data classification experiments, the results showed that in a small question samples with the label, the accuracy of question classification which using the label propagation algorithm is significantly improved.(2) On the basis of the known question on the category, to build a user model. The user model first consider user’s interest of the dynamic. It refers to the user in the new problems have been proposed for the question category of degree of interest, through the history of the user to answer calculate every question’s time weighting under the category, the greater the weighting sum of time indicates the user is more interested in questions under the category in the recent period. However, due to the ability of the user to answer questions vary, the user which interest degree is high may not give the answer to the high quality, so consider the degree of expertise of the user. According to the community question answering system adopted mechanism, building user link relation graph with weights, setting the community question answering system’s personalized into link algorithm with weight, thereby calculate the user degree of expertise.(3) It presents a question recommended methods based on user model. According to the user model building, due to construct user model reflects the user dynamic degree of interest and the user degree of expertise, both the value of the multiplication obtains the user and the new problems of matching values, constitute a user recommendation list based on the level of matching value, thereby completing recommendations for new questions. Use "Baidu Knows" crawl on data recommendation experiments, the results showed that using the recommended method recommended in the precision values of new questions (P@N-percent) has significant improvement.