A Review on Deep Learning Techniques Applied to Answer Selection

Proceedings of the 27th International Conference on Computational Linguistics, pages 2132–2144 Santa Fe, New Mexico, USA, August 20-26, 2018. 2132 A Review on Deep Learning Techniques Applied to Answer Selection Tuan Manh Lai Adobe Research tlai@adobe.com Trung Bui Adobe Research bui@adobe.com Sheng Li Adobe Research sheli@adobe.com Abstract Given a question and a set of candidate answers, answer selection is the task of identifying which of the candidates answers the question correctly. It is an important problem in natural language processing, with applications in many areas. Recently, many deep learning based methods have been proposed for the task. They produce impressive performance without relying on any feature engineering or expensive external resources. In this paper, we aim to provide a comprehensive review on deep learning methods applied to answer selection. 1 Introduction Answer selection is an active research field and has drawn a lot of attention from the natural language processing community. Given a question and a set of candidate answers, the task is to identify which of the candidates contains the correct answer to the question. For example, given the question “Who established the Nobel Prize?” and the following candidate answers: 1. The Nobel Prize was established more than 100 years ago. 2. The Fields Medal, established in 1936, is often described as the Nobel Prize of mathematics. 3. The Nobel Prize was established in the will of Alfred Nobel. The third answer should be selected. This example shows that simple word matching is not enough. Even though, all of the sentences contain the keywords “established” and “Nobel Prize”, only the third sentence answers the question. From this point, we assume that each candidate answer is a sentence although the discussion is also applicable to more general case such as each candidate answer is a passage. Answer selection is an important problem in its own right as well as in the context of open domain question answering. Although details vary from system to system, a typical open domain question an- swering system can be considered as consisting of: (a) question analysis (b) retrieval of potentially relevant documents (c) ranking and selecting of the most promising sentences (or more generally, pas- sages) within the retrieved documents; and optionally d) extracting the exact natural language phrase that answers the question (Prager, 2006; Ferrucci, 2012; Sequiera et al., 2017). Figure 1 depicts a typical question answering pipeline. In this setup, answer selection can be applied to identify the sentences that are most relevant to the question within the retrieved documents. Besides its application in open domain question answering, the techniques developed for answer selection can be potentially used to predict answer quality in community question answering (cQA) sites (Nakov et al., 2015). Previous work on answer selection typically relies on feature engineering, linguistic tools, or external resources (Wang et al., 2007; Wang and Manning, 2010; Heilman and Smith, 2010; Yih et al., 2013; Yao et al., 2013). Recently, many deep learning based methods have been proposed for the task (Bian et al., 2017; …

Content truncated to 500 words