Leveraging 2-hop Distant Supervision from Table Entity Pairs for Relation Extraction

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 410–420, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 410 Leveraging 2-hop Distant Supervision from Table Entity Pairs for Relation Extraction Xiang Deng The Ohio State University deng.595@buckeyemail.osu.edu Huan Sun The Ohio State University sun.397@osu.edu Abstract Distant supervision (DS) has been widely used to automatically construct (noisy) labeled data for relation extraction (RE). Given two enti- ties, distant supervision exploits sentences that directly mention them for predicting their se- mantic relation. We refer to this strategy as 1-hop DS, which unfortunately may not work well for long-tail entities with few support- ing sentences. In this paper, we introduce a new strategy named 2-hop DS to enhance dis- tantly supervised RE, based on the observation that there exist a large number of relational ta- bles on the Web which contain entity pairs that share common relations. We refer to such en- tity pairs as anchors for each other, and col- lect all sentences that mention the anchor en- tity pairs of a given target entity pair to help relation prediction. We develop a new neu- ral RE method REDS2 in the multi-instance learning paradigm, which adopts a hierarchi- cal model structure to fuse information respec- tively from 1-hop DS and 2-hop DS. Extensive experimental results on a benchmark dataset show that REDS2 can consistently outperform various baselines across different settings by a substantial margin.1 1 Introduction Relation extraction (RE) aims to extract semantic relations between two entities from unstructured text and is an important task in natural language processing (NLP). Formally, given an entity pair (e1, e2) from a knowledge base (KB) and a sen- tence (instance) that mentions them, RE tries to predict if a relation r from a predefined relation set exists between e1 and e2. A special relation NA is used if none of the predefined relations holds. Given that it is costly to construct large-scale labeled instances for RE, distant supervision (DS) 1Our source code and datasets are at https:// github.com/sunlab-osu/REDS2. Figure 1: Illustration of 2-hop distant supervision. The top panel shows a target entity pair, one sentence that mentions it, and the relation under study which cannot be inferred from the sentence. The middle gives part of a table from Wikipedia page “Mr. Basketball USA”, where we can extract anchors for the target entity pair. The bottom shows some sentences that are associated with the anchors, which more clearly indicate the un- derinvestigated relation and can be utilized to extract relations between the target entity pair. has been a popular strategy to automatically con- struct (noisy) training data. It assumes that if two entities hold a relation in a KB, all sentences men- tioning them express the same relation. Noticing that the DS assumption does not always hold and has the wrong labeling problem, many efforts in- cluding (Riedel et al., 2010; Hoffmann et al., 2011; Surdeanu et al., 2012) …

Content truncated to 500 words