Identifying Authoritative and Reliable Contents in Community Question Answering with Domain Knowledge

摘要

Community Question Answering (CQA) has emerged as a popular forum for users to ask and answer questions. Over the last few years, CQA portals such as Yahoo answers and Baidu Zhidao have exploded in popularity, and now provide a viable alternative to general purpose Web search. A number of answers submitted to address questions on CQA sites compose a valuable knowledge repository, which could be a gold mine for information retrieval as well as text mining. Two important questions in CQA research are focused on the quality of contents and the reputation of the answerers. Previous approaches for retrieving relevant and high quality content have been proposed, but not much work has been done on providing an integrated framework to solve these two problems. Besides, no research work has used both text and link information in their methods via leveraging existing ratings of answers and questions. In this paper, we present a novel approach to analyze questions and answers based on the topic modeling framework with Dirichlet forest priors (LDA-DF). We utilize information obtained from LDA-DF to construct a joint topical and link model to identify authorities and reliable answers on a CQA site.We evaluate our methods in a dataset obtained from Yahoo! Answers. With the new representation of topical structures on CQA datasets, using a limited amount of web resource, we show significant improvements over the state-of-art methods LDA-DF, LDA, and HLDA on performance of authority identification and answer ranking.