Hybrid Question Answering over Knowledge Base and Free Text

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 2397–2407, Osaka, Japan, December 11-17 2016. Hybrid Question Answering over Knowledge Base and Free Text Kun Xu1, Yansong Feng1,∗, Songfang Huang2 and Dongyan Zhao1 1Institute of Computer Science & Technology, Peking University, Beijing, China 2IBM China Research Lab, Beijing, China {xukun, fengyansong, zhaody}@pku.edu.cn huangsf@cn.ibm.com Abstract Recent trend in question answering (QA) systems focuses on using structured knowledge bases (KBs) to find answers. While these systems are able to provide more precise answers than infor- mation retrieval (IR) based QA systems, the natural incompleteness of KB inevitably limits the question scope that the system can answer. In this paper, we present a hybrid question answer- ing (hybrid-QA) system which exploits both structured knowledge base and free text to answer a question. The main challenge is to recognize the meaning of a question using these two resources, i.e., structured KB and free text. To address this, we map relational phrases to KB predicates and textual relations simultaneously, and further develop an integer linear program (ILP) model to infer on these candidates and provide a globally optimal solution. Experiments on benchmark datasets show that our system can benefit from both structured KB and free text, outperforming the state-of-the-art systems. 1 Introduction Recently, with the emergence of large structured knowledge bases (KBs) like DBpedia (Auer et al., 2007), Freebase (Bollacker et al., 2008) and Yago (Suchanek et al., 2007), increasing research efforts on automatically answering natural language questions has shifted from using text corpora only to large scale structured KBs like DBpedia, Freebase (known as KB-QA). Compared to pure text resources used in IR-based QA systems, structured knowledge bases may help to provide users with more accurate and concise answers, especially for factoid questions. Generally, the traditional KB-QA paradigm assumes that world knowledge can be encoded using a closed vocabulary of formal predicates. In this paradigm, the system is given a knowledge base as input, and the question answering problem reduces to semantic parsing, i.e., mapping from text to logical forms containing the predicates from the given knowledge base. However, the closed predicate vocabulary as- sumed by the traditional KB-QA paradigm has inherent limitations. First, a closed predicate vocabulary has limited coverage, as such vocabularies are typically powered by community efforts. Second, a closed predicate vocabulary may abstract away potentially relevant semantic differences. Third, even a logical form was produced, the answers may be incomplete due to the imperfection of the KB, which has been addressed by (Riedel et al., 2013; Chen et al., 2014). For example, no logical form could be produced for the question who is the front man of the band that wrote Coffee & TV. Because the semantics of front man cannot be adequately encoded using Freebase or DBpedia predicates. On the other hand, knowledge bases like DBpedia capture real world facts, and web resources like Wikipedia may provide a large repository of sentences that complement those facts. For instance, we can find in Wikipedia a …

Content truncated to 500 words