CaEXR: a joint extraction framework for causal relationships based on word-pair network
Files
(Published version)
Date
2024
Authors
Liu, C.
Fang, W.
Cheng, D.
Zhai, R.
Qin, L.
Editors
Huang, D.-S.
Si, Z.
Zhang, C.
Si, Z.
Zhang, C.
Advisors
Journal Title
Journal ISSN
Volume Title
Type:
Conference paper
Citation
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 2024 / Huang, D.-S., Si, Z., Zhang, C. (ed./s), vol.4, pp.446-458
Statement of Responsibility
Conference Name
20th International Conference, ICIC 2024 (5 Aug 2024 - 8 Aug 2024 : Tianjin, China)
Abstract
Extracting causal relationships from natural language texts is a fundamental work in constructing eventic graphs. Most previous existing studies are based on the linear representation models of syntactic dependency trees. However, these studies perform poorly for causal relationships extraction in complex contexts, due to the diffcultity of constructing contextual causal information interaction channels, which are crucial for models to understand causal relationships.
In addition, existing syntactic dependency tree based models ignore the role of dependency connectivity of causal connectives in syntactic features, which leads to the limited ability of these models to characterise long-distance explicit causal sentences. To address these two issues, we propose a novel joint model named CaEXR. We first reorganize syntactic dependency trees into causal dependency trees according to the causal connectives, which strengthens the long-distance explicit causal features.
Then, we transform linear sequence features into high-dimensional word-pair grid features based on a learnable conditional layer normalization module to enchance contextual causal interaction information. Experimental results show that our method achieves the best results on different levels of causal relationships extraction tasks.
School/Discipline
Dissertation Note
Provenance
Description
Access Status
Rights
Copyright 2024 The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.
Access Condition Notes: Accepted manuscript available after 1 October 2025