Deep human answer understanding for natural reverse QA

摘要

This study focuses on a reverse question answering (QA) procedure, in which machines proactively raise questions and humans supply the answers. This procedure exists in many real human–machine interaction applications. However, a crucial problem in human–machine interaction is answer understanding. Existing solutions have relied on mandatory option term selections to avoid automatic answer understanding. However, these solutions have led to unnatural human–computer interaction and negatively affected user experience. Thus, we propose a novel deep answer understanding network, AntNet, for reverse QA. The network consists of three new modules, namely, a skeleton attention for questions, a relevance-aware representation of answers, and a multi-hop-based fusion. Furthermore, to alleviate the negative influences of some quite difficult human answers, an improved self-paced learning strategy is proposed to train the AntNet by assigning different weights to training samples according to their learning difficulties. Given that answer understanding for reverse QA has not been explored, a new data corpus is compiled in this study. Experimental results indicate that our proposed network is significantly better than existing methods and those modified from classical natural language processing deep models. The effectiveness of the three modules and the improved self-paced learning strategy is also verified.