ChatCEE: a causality extraction framework based on large language model

Liu, Bozhen

doi:10.1007/s10799-026-00484-7

ChatCEE: a causality extraction framework based on large language model

2026-02-01

0

PRE

AI

DOI：10.1007/s10799-026-00484-7

Original

Original request for help

Save

Abstract

En 中文

Large language models (LLMs), exemplified by ChatGPT and DeepSeek, have demonstrated powerful language understanding and text generation capabilities, opening up new possibilities to address the traditional problem of causality extraction. This paper focuses on the core task of causality extraction and adopts task-planning techniques in prompt engineering to develop a two-stage framework based on LLMs, known as ChatCEE (chat for cause-effect extraction), for automatic causality extraction with certain interpretable ability. The ChatCEE framework simplifies the extraction process by utilizing clauses and significantly enhances the performance of LLMs through the use of extraction templates and one correct extraction example. To validate the effectiveness of the framework, a comparison experiment with four types of prompts was conducted on two datasets: the hazardous chemical accident investigation report dataset and the CEC public dataset, using five LLMs, including GPT-4o and DeepSeek. The results showed that ChatCEE achieved F1 scores of 86.12% and 91.81% (GPT-4o) on the two datasets, outperforming existing model results. This framework achieves interpretable, automated causality extraction that is both cost-effective and efficient. The research presented in this paper offers a novel perspective on the problem of causality extraction, with processes and methods that can be applied to text understanding, question answering systems, and eventic graph (EG) construction.

Keywords:

Large language models

Prompt engineering

Causality extraction

ChatCEE

ChatCEE: a causality extraction framework based on large language model

Abstract

Journal

Organization

Cited Papers