Iterative Document-level Information Extraction via Imitation Learning

Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 1858–1874 May 2-6, 2023 ©2023 Association for Computational Linguistics Iterative Document-level Information Extraction via Imitation Learning Yunmo Chen1 William Gantt2 Weiwei Gu2 Tongfei Chen3 Aaron Steven White2 Benjamin Van Durme1 1 Johns Hopkins University 2 University of Rochester 3 Microsoft Semantic Machines {yunmo|vandurme}@jhu.edu, {wgantt@ur.|wgu7@ur.|aaron.white@}rochester.edu, tongfei@pm.me Abstract We present a novel iterative extraction model, ITERX, for extracting complex relations, or templates, i.e., #-tuples representing a map- ping from named slots to spans of text within a document. Documents may feature zero or more instances of a template of any given type, and the task of template extraction entails iden- tifying the templates in a document and ex- tracting each template’s slot values. Our imita- tion learning approach casts the problem as a Markov decision process (MDP), and relieves the need to use predefined template orders to train an extractor. It leads to state-of-the-art results on two established benchmarks – 4-ary relation extraction on SCIREX and template extraction on MUC-4 – as well as a strong baseline on the new BETTER Granular task.1 1 Introduction A variety of tasks in information extraction (IE) require synthesizing information across multiple sentences, up to the length of an entire document. The centrality of document-level reasoning to IE has been underscored by an intense research fo- cus in recent years on problems such as argument linking (Ebner et al., 2020; Li et al., 2021, i.a.), #-ary relation extraction (Quirk and Poon, 2017; Yao et al., 2019; Jain et al., 2020, i.a.), and — our primary focus — template extraction (Du et al., 2021b; Huang et al., 2021, i.a.). Construed broadly, template extraction is general enough to subsume certain other document-level extraction tasks, including #-ary relation extrac- tion. Motivated by this consideration, we propose to treat these problems under a unified framework of generalized template extraction (§2).2 Figure 1 shows 4-ary relations from the SCIREX dataset (Jain et al., 2020), presented as simple templates. 1 Code available at github.com/wanmok/iterx. 2 We encourage the reader to consult Appendix A for a discussion of some important differences between generalized template extraction and traditional event extraction. In this paper, we propose a Multi-sentiment-resource Enhanced Attention Network (MEAN)… Despite the remarkable progress made by the previous work, we argue that sentiment analysis still remains a challenge… Movie Review (MR) and Stanford Sentiment Treebank (SST) are used to evaluate our model… We adopt classification accuracy as the evaluation metric. Template 1 Task Method Dataset Metric sentiment analysis MEAN Movie Review accuracy Template 2 Task Method Dataset Metric sentiment analysis MEAN SST accuracy Figure 1: An example of multi-template extraction on a document (an NLP paper; Lei et al. (2018)) from the SCIREX dataset. An agent reads the entire paper and iteratively generates templates, each consisting of slots for Task, Method, Dataset, and Metric. Since documents typically describe multiple complex events and relations, template extraction systems must be capable of predicting multiple templates per document. Existing approaches …

Content truncated to 500 words