Research On Expert Recommendation In Community Question Answering

Community Question Answer system plays an increasingly important role in the Internet age.The proportion of using the system is getting higher and higher for the Internet users to obtain knowledge and solve problems.While the growing number of Community Question Answer system users brings more wealth of knowledge,it also brings a series of problems that affecting the user experience.For example,many questions cannot be timely recommend to proper responders due to the increasing of askers,therefore the asker has to wait for a long time for the best answer;many questions are lack of accurate answers and therefore is hard for other users to locate the information they want when searching.Aiming at the issues of the large number of Community Question Answer users and the lack of content words in questions and answers,this paper proposes a hybrid expert finding model based on topic analysis and link analysis.Firstly,it uses topic analysis to distill the users list that has higher similarity with the new question topic;then,the link analysis is used to rank the users in the candidate list.Before extracting the latent topics of questions and answers in Community Question Answering,the Tag-Topic Model used in this model extends semantic in questions tags through Wikipedia data,so as to solve the short text issue existing in the Community Question Answering.Through the combination of word occurrence and tag occurrence for theme confirming,Tag-Topic Model can increase the accuracy of latent topic information extraction of questions and answers in topic analysis model.Based on the topic sensitive PageRank algorithm,the link analysis method used in hybrid expert finding model is topic sensibility expert ranking algorithm,that is,to calculate the user authority consider of the user relationship network and users' feedback information.The recommendation index between users and new questions is calculated based on the combination of users' interests and authority.To verify the actual performance of of hybrid expert recommendation model,this paper has done enough experiment in the dataset of Stack Overflow,and made the performance comparison through three different evaluation metrics(perplexity,nDCG@1,nDCG@3,nDCG@5)with four existing Community Question Answer expert recommendation algorithms.The experiment result shows that the hybrid expert recommendation model increases 3.67%~9.40% in nDCG@N evaluation index and 8.62% in MAP metrics by comparing with best method existed.