Knowing False Negatives: An Adversarial Training Method for Distantly Supervised Relation Extraction

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 9661–9672 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 9661 Knowing False Negatives: An Adversarial Training Method for Distantly Supervised Relation Extraction Kailong Hao† Botao Yu† Wei Hu†, ‡, ∗ † State Key Laboratory for Novel Software Technology, Nanjing University, China ‡ National Institute of Healthcare Data Science, Nanjing University, China klhao.nju@gmail.com, btyu.nju@gmail.com, whu@nju.edu.cn Abstract Distantly supervised relation extraction (RE) automatically aligns unstructured text with re- lation instances in a knowledge base (KB). Due to the incompleteness of current KBs, sen- tences implying certain relations may be anno- tated as N/A instances, which causes the so- called false negative (FN) problem. Current RE methods usually overlook this problem, in- ducing improper biases in both training and testing procedures. To address this issue, we propose a two-stage approach. First, it finds out possible FN samples by heuristically lever- aging the memory mechanism of deep neural networks. Then, it aligns those unlabeled data with the training data into a unified feature space by adversarial training to assign pseudo labels and further utilize the information con- tained in them. Experiments on two wildly- used benchmark datasets demonstrate the ef- fectiveness of our approach. 1 Introduction Relation extraction (RE), defined as the task of identifying relations from unstructured text given two entity mentions, is a key component in many NLP applications such as knowledge base (KB) population (Ji and Grishman, 2011; De Sa et al., 2016) and question answering (Yu et al., 2017). Supervised RE requires a large amount of human labeled data, which is often labor-intensive and time-consuming. Distant supervision (DS), pro- posed by (Mintz et al., 2009), deals with this prob- lem by aligning a large corpus with a KB such as Freebase to provide weak supervision. It relies on the assumption that if two entities participate in a relation in a KB, then any sentence containing this entity pair expresses this relation. However, this assumption is too strong and brings in noises, particularly when the training KB is an external information source and not primar- ily derived from the training corpus. To alleviate ∗ Corresponding author Sentences Relations [Romano Mussolini] , the Italian jazz pi- anist , ... , died on Tuesday in [Rome] . death_place ... in the [Uptown] neighborhood of [New Orleans] , ... , where it was founded . neighborhood Table 1: False negative examples excerpted from the NYT10 (Riedel et al., 2010) N/A set. this problem, Riedel et al. (2010) relax this as- sumption to if two entities participate in a relation, at-least-one sentence that mentions these two enti- ties expresses this relation. Bunescu and Mooney (2007) propose multi-instance learning by organiz- ing sentences with the same entity pair into one set, referred to as a bag. Following this setting, Riedel et al. (2010); Hoffmann et al. (2011); Surdeanu et al. (2012) propose diverse hand-crafted features. Zeng et al. (2014, 2015) leverage convolutional neural networks (CNN) to learn the representations of instances. Lin et …

Content truncated to 500 words