Estimating Domain-Specific User Expertise for Answer Retrieval in Community Question-Answering Platforms

Abstract

Community Question-Answering (CQA) platforms leverage the inherent wisdom of the crowd - enabling users to retrieve quality information from domain experts through natural language. An important and challenging task is to identify reliable and trusted experts on large popular CQA platforms. State-of-the-art graph-based approaches to expertise estimation consider only user-user interactions without taking the relative contribution of individual answers into account, while pairwise-comparison approaches consider only pairs involving the best-answerer of each question. This research argues that there is a need to account for the user's relative contribution towards solving the question when estimating user expertise and proposes a content-agnostic measure of user contributions. This addition is incorporated into a competition-based approach for ranking users' question answering ability. The paper analyses how improvements in user expertise estimation impact on applications in expert search and answer quality prediction. Experiments using the Yahoo! Chiebukuro data show encouraging performance improvements and robustness over state-of-the-art approaches.