Finding Experts In Community Question Answering

With the continuous development of Web2.0, the interaction between users becomes an emerging trend of the network. In order to offer convenient to users in community question answering system, the system must have a huge knowledge database and provided timely responses to newly submitted questions. In community-based Q&A system, submitted questions will be answered by other users which make how to finding suitable users for new questions becomes a new important spot. Newly submitted question can have promptly and accurately answer and the knowledge will be expanded then if we identify experts among massive users. So, finding expert in community-based Q&A system is very important and it is an effective way to development the system rapidly. At present, common methods to finding experts are topic model and link analysis.In this paper, firstly, we introduce the research background and related knowledge of expert finding in community-based Q&A system and then introduce topic model and link analysis. Secondly, we present a method to finding expert based on category participation. This method calculates the original expert score of user in each category and the similarity between the categories first and then gets the category participation of user with every category.The overall expert score of the category is a linear combination of original expert score in this category and participation score of other similar categories. Finally, we make some improvements to the original PageRank algorithm. We add weight to the direct edges between users according to the quality of answer.This paper present two method to finding expert in community-based Q&A system by analyzing the user link relationships, exploring the theme distribution of question categories and users and using user feedback behavior to evaluate the quality of the answers. In category participation-based method the user’s contribution in similar question categories can strengthen the user’s expert score in given question category. In weight PageRank algorithm adding weight to users’direct edges depending on the quality of answers.The data sets used in the experiments are collected from Yahoo! Answers. To evaluate the performance of the two methods, we compare our methods with several classical methods. To evaluate the performance for expert finding, we used two widely-used metrics MAP and AP@10. Higher score, better performance.The experiment results tell that our method shows better performance.