Corpus-level Fine-grained Entity Typing Using Contextual Information

Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 715–725, Lisbon, Portugal, 17-21 September 2015. c⃝2015 Association for Computational Linguistics. Corpus-level Fine-grained Entity Typing Using Contextual Information Yadollah Yaghoobzadeh and Hinrich Sch¨utze Center for Information and Language Processing University of Munich, Germany yadollah@cis.lmu.de Abstract This paper addresses the problem of corpus-level entity typing, i.e., inferring from a large corpus that an entity is a member of a class such as “food” or “artist”. The application of entity typ- ing we are interested in is knowledge base completion, specifically, to learn which classes an entity is a member of. We pro- pose FIGMENT to tackle this problem. FIGMENT is embedding-based and com- bines (i) a global model that scores based on aggregated contextual information of an entity and (ii) a context model that first scores the individual occurrences of an en- tity and then aggregates the scores. In our evaluation, FIGMENT strongly out- performs an approach to entity typing that relies on relations obtained by an open in- formation extraction system. 1 Introduction Natural language understanding (NLU) is not pos- sible without knowledge about the world – partly so because world knowledge is needed for many NLP tasks that must be addressed as part of NLU; e.g., many coreference ambiguities can only be re- solved based on world knowledge. It is also true because most NLU applications combine a vari- ety of information sources that include both text sources and knowledge bases; e.g., question an- swering systems need access to knowledge bases like gazetteers. Thus, high-quality knowledge bases are critical for successful NLU. Unfortunately, most knowledge bases are in- complete. The effort required to create knowledge bases is considerable and since the world changes, it will always continue. Knowledge bases are therefore always in need of updates and correc- tions. To address this problem, we present an in- formation extraction method that can be used for knowledge base completion. In contrast to most other work on knowledge base completion, we fo- cus on fine-grained classification of entities as op- posed to relations between entities. The goal of knowledge base completion is to acquire knowledge in general as opposed to de- tailed analysis of an individual context or sen- tence. Therefore, our approach is corpus-level: We infer the types of an entity by considering the set of all of its mentions in the corpus. In contrast, named entity recognition (NER) is context-level or sentence-level: NER infers the type of an entity in a particular context. As will be discussed in more detail in the following sections, the problems of corpus-level entity typing vs. context/sentence- level entity typing are quite different. This is partly because the objectives of optimizing ac- curacy on the context-level vs. optimizing accu- racy on the corpus-level are different and partly because evaluation measures for corpus-level and context-level entity typing are different. We define our problem as follows. Let K be a knowledge base that models a set E of entities, a set T of …

Content truncated to 500 words