Study On Best Answer Policies In Community-based Question Answering Services

Community-based Question and Answering (CQA) system has now become an emerging Web 2.0 services. Unlike traditional automatic question answering system, after people putting forward their own questions, other users can utilize their own experience to answer the corresponding questions. After a period of time, askers can pick an answer of all the others'replies as the best answer, or select the best answer by popular voting. However, since the process is completely through the manual manipulation, there is no automatic machine evaluation, not only resulting in the decline of answer qualities in CQAsystem, but also made it difficult for other systems such as search engine to reuse the question/answer pair as the knowledge base. Therefore, we propose the research problem of mechanism of Best Answer Policies in CQA system.Currently, a question on CQA System can only have one best answer. However, through our observations, we found that not all questions only have unique answers. Many questions may have alternative answers. Therefore, we try to propose the answer summarization techniques which can generate summaries based on all answers of questions as the alternative of best answer. By doing this, questions can have complete and accurate answers so that it made up the pitfalls of current system and it also contributes a lot for other systems reuse as large Q&A archives.Specifically, we first carried out an in-depth analysis of questions and answers on CQA system. We made the answer taxonomy based on the criteria"whether the best answer can be used if similar questions are asked again"and made the question taxonomy based on the question intention. We found that among them more over 78% questions have reusable answers but no more than 48% questions have unique answers. Open-type questions and opinion-type questions occupy the most for having multiple good answers.Then, we propose question-type focused answer summarization algorithms for these two question types. For open-type questions, we propose answer clustering algorithm and cluster labeling algorithm. For sentiment-type of opinion-type questions, we propose sentiment polar type judging algorithm. For list-type of opinion-type questions, we propose sentence clustering algorithm. Moreover, we propose the criteria of information content and readability to judge the effect of generated summary. Experiment results show that our customized question-focused summarization techniques can improve CQA answer quality significantly.