Exploiting response patterns for identifying topical experts in StackOverflow

Abstract

The popularity of community question answer (CQA) forums like Stack Overflow, Yahoo Answers and Quora is increasing tremendously with thousands of questions being posted each day and about thrice the number of responses being provided. With such query explosion, users participating in these forums receive a huge number of postings that adversely affects their responsiveness and also the quality of the responses. Hence, identifying topical experts is necessary to improve the efficacy of these systems in terms of both response time and quality. Although expert detection in CQA forums has traditionally been a topic of wide interest, however, many of the proposed techniques use features set that reflect the popularity of the responses of the responder rather than the difficulty level of the questions being responded. In this paper we provide measures of labeling difficult questions and use the number of difficult questions responded by a user combined with other user interaction parameters in identifying potential topical experts. Using a random forest classifier with the proposed feature set on Stack Overflow data, we obtain an improvement in accuracy of 5–16% over existing techniques, in detecting topical experts.