Learning Dynamic Context Augmentation for Global Entity Linking

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 271–281, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 271 Learning Dynamic Context Augmentation for Global Entity Linking Xiyuan Yang1, Xiaotao Gu2, Sheng Lin1, Siliang Tang1∗, Yueting Zhuang1, Fei Wu1, Zhigang Chen3, Guoping Hu3 & Xiang Ren4 1Zhejiang University, 2University of Illinois at Urbana Champaign 3iFLYTEK Research, 4University of Southern California {yangxiyuan, shenglin, siliang, yzhuang, wufei}@zju.edu.cn, xiaotao2@illinois.edu, {zgchen, gphu}@iflytek.com, xiangren@usc.edu Abstract Despite of the recent success of collective en- tity linking (EL) methods, these “global” in- ference methods may yield sub-optimal results when the “all-mention coherence” assumption breaks, and often suffer from high computa- tional cost at the inference stage, due to the complex search space. In this paper, we pro- pose a simple yet effective solution, called Dynamic Context Augmentation (DCA), for collective EL, which requires only one pass through the mentions in a document. DCA sequentially accumulates context information to make efficient, collective inference, and can cope with different local EL models as a plug- and-enhance module. We explore both super- vised and reinforcement learning strategies for learning the DCA model. Extensive experi- ments1 show the effectiveness of our model with different learning settings, base models, decision orders and attention mechanisms. 1 Introduction Linking mentions of entities in text to knowledge base entries (i.e., entity linking, or EL) is critical to understanding and structuring text corpora. In general, EL is approached by first obtaining can- didate entities for each mention, and then identi- fying the true referent among the candidate enti- ties. Prior distribution and local contexts, either in the form of hand-crafted features (Ratinov et al., 2011; Shen et al., 2015) or dense embeddings (He et al., 2013; Nguyen et al., 2016; Francis-Landau et al., 2016), play key roles in distinguishing dif- ferent candidates. However, in many cases, local features can be too sparse to provide sufficient in- formation for disambiguation. To alleviate this problem, various collective EL models have been proposed to globally optimize ∗Corresponding author. 1Code and data is available at https://github.com/YoungXiyuan/DCA ��� ��������� � ��� ������ ������ ����� ���� ������ ������� ���� ���� ��� ����������� �� ���������� �� ��������� New Mexico ����������������������� Beirut ������� ������ ������� ������������������� Washington ���������������� ���������� ������������ �������������� ���� ��������������� ��������������� New Mexico ���������� ������������ � �������� � ���������������� � ����������� � ����������������� � ������������������������� Beirut ������ ������ ������ ������������������� Washington ����������� ���� ���������� ������������ � ���������� � �������� � �������������� � � � ����������� Mention ��������� ������ ��������� ������ Figure 1: An Illustration of the Dynamic Context Augmentation process. A traditional global EL model jointly optimizes the linking configuration after itera- tive calculations over all mentions, which is compu- tationally expensive. In contrast, the DCA process only requires one pass of the document to accumulate knowledge from previously linked mentions to enhance fast future inference. the inter-entity coherence between mentions in the same document (Hoffart et al., 2011; Cheng and Roth, 2013; Nguyen …

Content truncated to 500 words