Research On Key Techniques In Community Question Answering Site

Recently, with the rapid development and wide application of Internet, community question answering system has attracted a lot of attention and becomes an important and hot research topic in the areas of natural language processing and information retrieval. Community question answering system provides services, e.g. question search and question recommendations, in order to answer users’questions as soon as possible. However, existing solutions have the following three drawbacks. Firstly, most of the existing solutions are based on natural language model and simply use text retrivel technologies, which disregard the structural features and interactive information of community question answering system. Secondly, most of the existing solutions are not customized, since they do not consider the user’s language habits, interests, preferences and so on. At last, to determine the best answer of a given question, the existing solutions rely solely on asker or other voters, which may result in long response time. This paper thus proposes to solve the problems by a) providing customized services of question retrievel and question recommendation; b) giving a novel solution of choosing the best answer online. The contributions are summarized as follows:1. Propose a similar question search approach by using categorization information and dependency syntactic tree. The approach not only makes full use of the structural features of community question answering system, but also takes advantage of the text analysis technologies for natural language processing and considers the user’s history of questioning. Experiments show that:users’ questioning history and categorization information help reduce the search space significantly, and by applying dependency syntactic tree to extract keywords for search, the accuracy is improved.2. Propose a personalized question recommendation scheme. The scheme explores the structural features on community question answering systems by analyzing existing community question answering systesms like Yahoo! Answers. Based on the features, the scheme applies machine learning classification technologies, combining all users’ answering history to recommend questions. Experiments show that:the features that affect the satisfaction of recommendation are quite personal. The question recommendation scheme based on machine learning classification technologies achieves high precision and recall, and the performance is becoming better as users’answering history keeps growing.3. Propose a best answer selection method towards factoid question. The method doesn’t need training data or manual annotation. It selects the best answer automatically by making use of the similarity among the answers for factoid questions and the for/against information from voters. Experiments show that: the method can be used online, and has high precision.