Towards Robust Expert Finding in Community Question Answering Platforms

TL;DR

This work tackles Expert Finding in Community Question Answering by introducing TUEF, a topic-oriented, user-interaction model that fuses content and social information through a multi-layer graph. The approach combines topic identification, graph-based candidate exploration, and a LambdaMART ranking framework to transparently select and rank expert candidates for each question. Empirical results on StackOverflow show substantial gains over state-of-the-art baselines, with notable improvements in MRR and top-1 precision, driven by the integration of heterogeneous signals and effective exploration. The findings underscore the practical value of joint content-social modeling for robust, trustworthy routing of questions to qualified experts, with future work targeting interpretability and fairness.

Abstract

This paper introduces TUEF, a topic-oriented user-interaction model for fair Expert Finding in Community Question Answering (CQA) platforms. The Expert Finding task in CQA platforms involves identifying proficient users capable of providing accurate answers to questions from the community. To this aim, TUEF improves the robustness and credibility of the CQA platform through a more precise Expert Finding component. The key idea of TUEF is to exploit diverse types of information, specifically, content and social information, to identify more precisely experts thus improving the robustness of the task. We assess TUEF through reproducible experiments conducted on a large-scale dataset from StackOverflow. The results consistently demonstrate that TUEF outperforms state-of-the-art competitors while promoting transparent expert identification.

Figures (2)

• Figure 1: TUEF approach representation highlighting the distinct components.
• Figure 2: The feature importance as determined by the Learning-to-Rank approach.