Joint Representation Learning for Expert Discovery

Abstract

The success of community question answering (CQA) platforms is in part dependent on how well questions are matched with the best experts that can effectively answer them in a timely manner. As the complexity of questions on CQA platforms increases, necessitating the collaboration of multiple experts, existing expert-finding methods are inadequate because they are designed to evaluate the suitability of individual experts for specific questions rather than collaborative expertise. This paper introduces a novel approach to identify collaborative teams of experts by employing graph neural networks and kernel pooling techniques, trained end-to-end. Our approach not only predicts potential individual experts but also forms teams that collectively enhance the answer quality. Through extensive experiments on real-life datasets, we show our proposed model is able to (1) find more qualified experts for new questions by at least 4.4% and 6.7% superior ranks in terms of ranking metrics NDCG and MAP, respectively, compared to the best expert finding baselines; (2) form a collaborative team of experts with 4.7% higher skill coverage than collaborative expert finding baselines.

Morteza Zihayat

Principal Investigator

Dr. Morteza Zihayat is a Canada Research Chair (CRC) in Human-Centered AI and Associate Professor at Toronto Metropolitan University, Faculty of Engineering and Architectural Science. He also holds appointments as Adjunct Associate Professor at the University of Waterloo (Management Sciences) and IBM Faculty Fellow at IBM Centre for Advanced Studies. He is the Director of the Human-Centered Machine Intelligence Lab.