Skip to main navigation Skip to search Skip to main content

A Unified Information Extraction System Based on Role Recognition and Combination

  • Yadong Zhang*
  • , Man Lan
  • *Corresponding author for this work
  • East China Normal University

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

In this paper, we propose a unified information extraction system, which handles event extraction (EE) and relation extraction (RE) tasks. Given context and schema, event extraction aims to extract the events and the specific roles in the events, and relation extraction extracts all SPO triples. We formulate event extraction and relation extraction as one extraction schema, that is, role recognition and role combination. We use Multi-Label Pointer Network (MLPN) to recognize composite roles that contain both event/relation and role information and simultaneously train a Co-occurrence Matrix (CM) to determine the co-occurrence relationship of composite roles, i.e., whether two roles describe the same event/relation. Using such a Unified model based on Role Recognition and Combination (URRC) and corresponding combination strategy, we implement three tasks: sentence-level event extraction, document-level event extraction, and relation extraction. In LIC 2021, our model achieved 6th in the Multi-format Information Extraction racing track with an average F1 score of 77.44% in the final test dataset of three subtasks.

Original languageEnglish
Title of host publicationNatural Language Processing and Chinese Computing - 10th CCF International Conference, NLPCC 2021, Proceedings
EditorsLu Wang, Yansong Feng, Yu Hong, Ruifang He
PublisherSpringer Science and Business Media Deutschland GmbH
Pages447-459
Number of pages13
ISBN (Print)9783030884826
DOIs
StatePublished - 2021
Event10th CCF Conference on Natural Language Processing and Chinese Computing, NLPCC 2021 - Qingdao, China
Duration: 13 Oct 202117 Oct 2021

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13029 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference10th CCF Conference on Natural Language Processing and Chinese Computing, NLPCC 2021
Country/TerritoryChina
CityQingdao
Period13/10/2117/10/21

Keywords

  • Document-level event extraction
  • Event extraction
  • Relation extraction

Fingerprint

Dive into the research topics of 'A Unified Information Extraction System Based on Role Recognition and Combination'. Together they form a unique fingerprint.

Cite this