Attention-based combination of CNN and RNN for relation classification

  • Xiaoyu Guo
  • , Hui Zhang
  • , Rui Liu*
  • , Xin Ding
  • , Runqi Tian
  • , Bencheng Wang
  • *Corresponding author for this work

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

Abstract

Relation classification is an essential task in natural language processing (NLP) in order to extract structured data from sentences. In this paper, we propose a novel model Att-ComNN combining convolutional neural network (CNN) and bidirectional recurrent neural network (RNN) for relation classification. By combining RNN and CNN, we obtain more accurate context representations of words, which benefits classifying relations. Besides, with both shortest dependency path (SDP) attention and pooling attention added, this model captures the most informative context representation for better classification without using other handcrafted features. The results of experiments show that our model improves the relation classification performance on the SemEval-2010 Task 8 and outperforms most of previous state-of-the-art methods, including those depending on much richer forms of handcrafted features and prior knowledge.

Original languageEnglish
Title of host publicationNeural Information Processing - 25th International Conference, ICONIP 2018, Proceedings
EditorsSeiichi Ozawa, Andrew Chi Sing Leung, Long Cheng
PublisherSpringer Verlag
Pages244-255
Number of pages12
ISBN (Print)9783030042110
DOIs
StatePublished - 2018
Event25th International Conference on Neural Information Processing, ICONIP 2018 - Siem Reap, Cambodia
Duration: 13 Dec 201816 Dec 2018

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume11304 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference25th International Conference on Neural Information Processing, ICONIP 2018
Country/TerritoryCambodia
CitySiem Reap
Period13/12/1816/12/18

Keywords

  • Attention mechanism
  • Deep neural network
  • Relation classification

Fingerprint

Dive into the research topics of 'Attention-based combination of CNN and RNN for relation classification'. Together they form a unique fingerprint.

Cite this