Knowledge as A Bridge: Improving Cross-domain Answer Selection with External Knowledge

Proceedings of the 27th International Conference on Computational Linguistics, pages 3295–3305 Santa Fe, New Mexico, USA, August 20-26, 2018. 3295 Knowledge as A Bridge: Improving Cross-domain Answer Selection with External Knowledge Yang Deng1, Ying Shen1,∗, Min Yang2, Yaliang Li3, Nan Du3, Wei Fan3, Kai Lei1 1School of Electronics and Computer Engineering, Peking University Shenzhen Graduate School 2Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences 3Tencent Medical AI Lab ydeng@pku.edu.cn, shenying@pkusz.edu.cn, min.yang@siat.ac.cn, {yaliangli,ndu,davidwfan}@tencent.com, leik@pkusz.edu.cn Abstract Answer selection is an important but challenging task. Significant progress has been made in domains where a large amount of labeled training data is available. However, obtaining rich annotated data is a time-consuming and expensive process, creating a substantial barrier for ap- plying answer selection models to a new domain which has limited labeled data. In this paper, we propose Knowledge-aware Attentive Network (KAN), a transfer learning framework for cross- domain answer selection, which uses the knowledge base as a bridge to enable knowledge trans- fer from the source domain to the target domains. Specifically, we design a knowledge module to integrate the knowledge-based representational learning into answer selection models. The learned knowledge-based representations are shared by source and target domains, which not only leverages large amounts of cross-domain data, but also benefits from a regularization effect that leads to more general representations to help tasks in new domains. To verify the effec- tiveness of our model, we use SQuAD-T dataset as the source domain and three other datasets (i.e., Yahoo QA, TREC QA and InsuranceQA) as the target domains. The experimental results demonstrate that KAN has remarkable applicability and generality, and consistently outperforms the strong competitors by a noticeable margin for cross-domain answer selection. 1 Introduction Answer selection, which is a key component of question answering (QA), has attracted increasing at- tention recently due to its broad applications in natural language processing and information retrieval, such as factoid question answering (Wang et al., 2007) and community-based question answering (Tay et al., 2017). Given a question, answer selection aims to pick out the most relevant answer from a set of candidates. In the literature, answer selection has been extensively studied in the last decade (Severyn and Moschitti, 2015; Wang and Nyberg, 2015; Tan et al., 2016; dos Santos et al., 2016). Despite the effectiveness of previous studies, answer selection remains a challenge in real-world ap- plications for two reasons. (1) The background information and knowledge beyond the context, which play crucial roles in human text comprehension, have received little attention in recent work for answer selection. (2) Impressive answer selection performances were achieved in domains where a large amount of labeled data is available. However, such fruitful results are subject to an assumption that the test data should be drawn from the same distribution as the training data. Previous studies struggle to cope with answer selection across different data domains. For example, the model trained on SQuAD-T dataset that consists of open-domain factoid questions, is difficult to generalize to the InsuranceQA dataset that consists of …

Content truncated to 500 words