Integrating Subject, Type, and Property Identification for Simple Question Answering over Knowledge Base

Proceedings of the The 8th International Joint Conference on Natural Language Processing, pages 976–985, Taipei, Taiwan, November 27 – December 1, 2017 c⃝2017 AFNLP Integrating Subject, Type, and Property Identification for Simple Question Answering over Knowledge Base Wei-Chuan Hsiao, Hen-Hsen Huang and Hsin-Hsi Chen Department of Computer Science and Information Engineering National Taiwan University, Taipei, Taiwan weichuanhsiao@gmail.com; hhhuang@nlg.csie.ntu.edu.tw; hhchen@ntu.edu.tw Abstract This paper presents an approach to identify subject, type and property from knowledge base for answering simple questions. We propose new fea- tures to rank entity candidates in KB. Besides, we split a relation in KB into type and property. Each of them is modeled by a bi-directional LSTM. Experimental results show that our model achieves the state-of-the-art per- formance on the SimpleQuestions da- taset. The hard questions in the experi- ments are also analyzed in detail. 1 Introduction With the popularity of the Internet, more and more new information is generated every day. The in- formation may be stored in unstructured data, such as Wikipedia, which is presented as an article or web page, or in the form of structured data. Knowledge base (KB), such as Freebase (Bol- lacker et al., 2008) and DBpedia (Lehmann et al., 2015), is very popular. The information in the KB is a description of the relationship between two en- tities and is stored in the form of (subject, relation, object) triple. In KB, each entity is represented by a unique id, for example, J.K. Rowling's mid (Ma- chine ID) in Freebase is “m.042xh”. With these large-scale open-domain KBs, it is important to access the knowledge efficiently and effectively to meet what users need. The most di- rect and close to people's life is question answering (QA) system in natural language. People can ask any questions in their familiar languages, and then use the QA system to get answers from the web re- sources. Current QA research is often based on KB to find appropriate triples for answering question, which is called QA over KB. QA systems can be classified by the form of the question. There are two categories of questions in QA system, i.e., simple questions and complex questions. Simple questions can be answered by exactly one triple in the KB. For example, the question “Who is the author of Harry Potter?” can be answered by triple “(Harry Potter, author, J.K. Rowling)”. Although this category is “simple” question, retrieving a triple from the KB is not a trivial task due to the billions of facts in the KB. Complex questions contain more restrictions. These questions may involve two or more triples in the KB, or have other semantic constraints to re- strict the answers to a smaller set. For example, “the first” in the question “What is the name of the first Harry Potter novel?” restricts that there is only one answer. Previous researches showed that sim- ple questions are the more common category in community QA websites (Fader et al., 2013). This paper focuses on factoid simple question-answer- ing over Freebase. …

Content truncated to 500 words