Title: A cascade decoder-based method for complex causal relation extraction

Authors: Wanli Su; Bo Shen

Addresses: School of Electronic and Information Engineering, Beijing Jiaotong University, Beijing, 100044, China ' Key Laboratory of Communication and Information Systems, Beijing Municipal Commission of Education, Beijing, 100044, China

Abstract: Causal relation extraction is an important task in textual causal analysis, aiming to extract causal relation pairs from text. Although existing models have achieved notable success, they primarily focus on single causal relations in text and still exhibit limitations when dealing with complex causal contexts. To address this issue, we propose a cascade decoder network (CDN) based on a causal mapping mechanism. This model formulates causal relations as a directional mapping process from causes to effects, resolving complex causal structures through a two-stage decoding framework: the first-stage decoder identifies potential cause entities from the text, while the second-stage decoder maps specific causes to their corresponding effect entities. To tackle the challenge of long-distance dependencies, the model integrates multi-scale feature extraction and an interactive attention mechanism to dynamically capture semantic associations between causal entities. Experimental results demonstrate that our method outperforms baseline models on public causal extraction datasets, achieving higher F1 scores under strict evaluation criteria.

Keywords: complex causal extraction; cause-effect mapping; entity identification; cross-attention; cascade decoder.

DOI: 10.1504/IJAISC.2025.151481

International Journal of Artificial Intelligence and Soft Computing, 2025 Vol.9 No.4, pp.360 - 378

Received: 23 May 2025
Accepted: 11 Nov 2025
Published online: 02 Feb 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article