Ranking answers of comparative questions using heterogeneous information organization from social media

Abstract

One of the most typical subjective questions in CQA Web sites is comparative question. Comparative question answering is quite different from the factual question answering. First, it aims to generate opinions, conclusions, reasons, etc, rather than only facts. Second, the answers are usually summarized from different domains (or aspects). Despite the success of the previous answer ranking approaches, none of them can be adaptive well to ranking comparative question answers. They overlook the external knowledge as semantic expansions, mining the related domain terms of the questions as clues for ranking comparative question answers. In this paper, we employ the heterogeneous information organization technique as the external knowledge generator and propose a domain space model-based ranking scheme to integrate the domain terms in questions for answer re-ranking. Through empirical comparisons, the proposed approach significantly outperforms the state-of-the-art (SOTA) approaches. Meanwhile, we also verify that our proposed approach is robust to the noise in domain term set.