Bayesian Deep Learning for Interactive Community Question Answering

Abstract

Human-in-the-loop interactive learning has been shown to be effective for best solution selection tasks. Bayesian Optimisation (BO) reduces the amount of user interaction required but has so far relied on shallow models rather than end-to-end deep learning. This paper leverages recent advances in Bayesian deep learning (BDL) to more accurately identify the best solution from a few rounds of interaction. We apply our approach to community question answering (cQA), finding that our BDL approach significantly outperforms existing methods while remaining robust to noise in the user feedback.

Tasks

Reproductions