Skip to main navigation Skip to search Skip to main content

M-HGN: Multi-information Enhanced Heterogeneous Graph Network for Multi-party Dialogue Reading Comprehension

  • Xiaoqian Gao
  • , Xiabing Zhou*
  • , Rui Cao
  • , Min Zhang
  • *Corresponding author for this work
  • Soochow University

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

Abstract

Multi-party dialogue reading comprehension aims to comprehend the dialogue context and answer the questions. Using explicit contextual information effectively and trying to capture the implicit information behind the dialogue lie at the heart of the task, which is still under-explored by previous works. In this paper, we propose a multi-information enhanced heterogeneous graph network to hierarchically model the dialogue context at different granularity levels and the multi-grained interactive relations among them. In detail, we construct a heterogeneous graph by regarding speakers, discourse structures, and questions as different types of nodes to model the information flows among them. In addition, we consider commonsense knowledge as a type of node in order to comprehend additional implicit information. Furthermore, we design a proper interaction module to incorporate these multi-dimensional features together. Experimental results show that our model achieves significant improvements on Molweni dataset. Also, extensive analysis indicates that our model can effectively capture multi-grained information and enhance the understanding of multi-party dialogue.

Original languageEnglish
Title of host publicationKnowledge Science, Engineering and Management - 17th International Conference, KSEM 2024, Proceedings
EditorsCungeng Cao, Huajun Chen, Liang Zhao, Junaid Arshad, Yonghao Wang, Taufiq Asyhari
PublisherSpringer Science and Business Media Deutschland GmbH
Pages371-383
Number of pages13
ISBN (Print)9789819754946
DOIs
StatePublished - 2024
Externally publishedYes
Event17th International Conference on Knowledge Science, Engineering and Management, KSEM 2024 - Birmingham, United Kingdom
Duration: 16 Aug 202418 Aug 2024

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume14885 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference17th International Conference on Knowledge Science, Engineering and Management, KSEM 2024
Country/TerritoryUnited Kingdom
CityBirmingham
Period16/08/2418/08/24

Keywords

  • Commonsense knowledge
  • Heterogeneous graph network
  • Multi-party dialogue
  • Reading comprehension

Fingerprint

Dive into the research topics of 'M-HGN: Multi-information Enhanced Heterogeneous Graph Network for Multi-party Dialogue Reading Comprehension'. Together they form a unique fingerprint.

Cite this