Who have got answers?: growing the pool of answerers in a smart enterprise social QA system

摘要: On top of an enterprise social platform, we are building a smart QA system that automatically routes questions to suitable employees who willing, able, and ready provide answers. Due lack history (training data) start with, in this paper, present optimization-based approach recommends both top-matched active (seed) inactive (prospect) answerers for given question. Our includes three parts. First, it uses predictive model find top-ranked seed by their fitness, including ability willingness, answer Second, distance metric learning discover prospects most similar the seeds identified first step. Third, constraint-based balance selection two steps. As result, not only does our solution route users, but also engages users grow pool answerers. real-world experiments routed 114 684 people from 400,000+ included 641 (93.7%) achieved about 70% answering rate with 83% answers received lot/full confidence.