Fact Discovery from Knowledge Base via Facet Decomposition

Proceedings of NAACL-HLT 2019, pages 2892–2901 Minneapolis, Minnesota, June 2 - June 7, 2019. c⃝2019 Association for Computational Linguistics 2892 Fact Discovery from Knowledge Base via Facet Decomposition Zihao Fu1, Yankai Lin2, Zhiyuan Liu2∗, Wai Lam1 1 Department of Systems Engineering and Engineering Management The Chinese University of Hong Kong, Hong Kong 2 Department of Computer Science and Technology, State Key Lab on Intelligent Technology and Systems, National Lab for Information Science and Technology, Tsinghua University, Beijing, China Abstract During the past few decades, knowledge bases (KBs) have experienced rapid growth. Never- theless, most KBs still suffer from serious in- completion. Researchers proposed many tasks such as knowledge base completion and re- lation prediction to help build the representa- tion of KBs. However, there are some issues unsettled towards enriching the KBs. Knowl- edge base completion and relation prediction assume that we know two elements of the fact triples and we are going to predict the miss- ing one. This assumption is too restricted in practice and prevents it from discovering new facts directly. To address this issue, we pro- pose a new task, namely, fact discovery from knowledge base. This task only requires that we know the head entity and the goal is to discover facts associated with the head en- tity. To tackle this new problem, we propose a novel framework that decomposes the dis- covery problem into several facet discovery components. We also propose a novel auto- encoder based facet component to estimate some facets of the fact. Besides, we propose a feedback learning component to share the in- formation between each facet. We evaluate our framework using a benchmark dataset and the experimental results show that our framework achieves promising results. We also conduct extensive analysis of our framework in discov- ering different kinds of facts. The source code of this paper can be obtained from https: //github.com/thunlp/FFD. 1 Introduction Recent years have witnessed the emergence and growth of many large-scale knowledge bases (KBs) such as Freebase (Bollacker et al., 2008), DBpedia (Lehmann et al., 2015), YAGO (Suchanek et al., 2007) and Wikidata (Vrandeˇci´c ∗ Corresponding author: Zhiyuan Liu (li- uzy@tsinghua.edu.cn). Serena Williams BirthPlace Best Female Tennis Player Award AwardReceived Serena Williams BirthPlace Best Female Tennis Player Award AwardReceived Gender Nationality Tennis Player Profession Language USA Female English Saginaw Saginaw Figure 1: In the FDKB task, only the head entity is given. The relation and tail entity should be discovered simultaneously given the head entity. and Kr¨otzsch, 2014) to store facts of the real world. Most KBs typically organize the complex structured information about facts in the form of triples (head entity, relation, tail entity), e.g., (Bill Gates, CEOof, Microsoft Inc.). These KBs have been widely used in many AI and NLP tasks such as text analysis (Berant et al., 2013), question answering (Bordes et al., 2014a), and information retrieval (Hoffmann et al., 2011). The construction of these KBs is always an on- going process due to the endless growth of real- world facts. Hence, many …

Content truncated to 500 words