Fine-grained Entity Typing through Increased Discourse Context and Adaptive Classification Thresholds

Proceedings of the 7th Joint Conference on Lexical and Computational Semantics (*SEM), pages 173–179 New Orleans, June 5-6, 2018. c⃝2018 Association for Computational Linguistics Fine-grained Entity Typing through Increased Discourse Context and Adaptive Classification Thresholds Sheng Zhang Johns Hopkins University zsheng2@jhu.edu Kevin Duh Johns Hopkins University kevinduh@cs.jhu.edu Benjamin Van Durme Johns Hopkins University vandurme@cs.jhu.edu Abstract Fine-grained entity typing is the task of as- signing fine-grained semantic types to entity mentions. We propose a neural architecture which learns a distributional semantic repre- sentation that leverages a greater amount of se- mantic context – both document and sentence level information – than prior work. We find that additional context improves performance, with further improvements gained by utiliz- ing adaptive classification thresholds. Experi- ments show that our approach without reliance on hand-crafted features achieves the state-of- the-art results on three benchmark datasets. 1 Introduction Named entity typing is the task of detecting the type (e.g., person, location, or organization) of a named entity in natural language text. Entity type information has shown to be useful in natural language tasks such as question answering (Lee et al., 2006), knowledge-base population (Carl- son et al., 2010; Mitchell et al., 2015), and co- reference resolution (Recasens et al., 2013). Mo- tivated by its application to downstream tasks, re- cent work on entity typing has moved beyond stan- dard coarse types towards finer-grained semantic types with richer ontologies (Lee et al., 2006; Ling and Weld, 2012; Yosef et al., 2012; Gillick et al., 2014; Del Corro et al., 2015). Rather than as- suming an entity can be uniquely categorized into a single type, the task has been approached as a multi-label classification problem: e.g., in “... be- came a top seller ... Monopoly is played in 114 countries. ...” (Figure 1), “Monopoly” is consid- ered both a game as well as a product. The state-of-the-art approach (Shimaoka et al., 2017) for fine-grained entity typing employs an attentive neural architecture to learn representa- tions of the entity mention as well as its con- text. These representations are then combined Entity Encoder Sentence-level Context Encoder Document-level Context Encoder sentence-level context document-level context Attention Feature representation Type embeddings … Logistic Regression Adaptive Thresholds [ /game, /product ] Output types: Featurizer “… became a top seller … is played in 114 countries. …” Monopoly Figure 1: Neural architecture for predicting the types of entity mention “Monopoly” in the text “... became a top seller ... Monopoly is played in 114 countries. ...”. Part of document-level context is omitted. with hand-crafted features (e.g., lexical and syn- tactic features), and fed into a linear classifier with a fixed threshold. While this approach outper- forms previous approaches which only use sparse binary features (Ling and Weld, 2012; Gillick et al., 2014) or distributed representations (Yo- gatama et al., 2015), it has a few drawbacks: (1) the representations of left and right contexts are learnt independently, ignoring their mutual con- nection; (2) the attention on context is computed solely upon the context, considering no alignment …

Content truncated to 500 words