TY - GEN
T1 - Logical Data Augmentation for Debiased Zero-Shot Stance Detection
AU - Cheng, Yinghan
AU - Hao, Shufeng
AU - Shi, Chongyang
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2026.
PY - 2026
Y1 - 2026
N2 - We propose a novel knowledge enhancement zero-shot stance detection method, which employs LLMs to construct data augmentation based on Wikipedia’s background knowledge and uses Causal debiasing techniques to calibrate the LLM bias. First, the method retrieves, filters and summarizes relevant Wikipedia knowledge using topic modeling and LLMs, while enhancing text comprehension through LLM-based paraphrasing. Second, a logical Chain-of-Thought module is employed to generate coherent augmented data by deriving logical expressions from text-target pairs, in order to address the limitations, such as the lack of logical relevance and the insufficient generalization ability in data augmentation. Finally, the method introduces causal counterfactual debiasing theory with a calibration network to mitigate LLM biases while improving generalization. Extensive experimental results demonstrate that the proposed method achieves superior performance over the state-of-the-art baselines.
AB - We propose a novel knowledge enhancement zero-shot stance detection method, which employs LLMs to construct data augmentation based on Wikipedia’s background knowledge and uses Causal debiasing techniques to calibrate the LLM bias. First, the method retrieves, filters and summarizes relevant Wikipedia knowledge using topic modeling and LLMs, while enhancing text comprehension through LLM-based paraphrasing. Second, a logical Chain-of-Thought module is employed to generate coherent augmented data by deriving logical expressions from text-target pairs, in order to address the limitations, such as the lack of logical relevance and the insufficient generalization ability in data augmentation. Finally, the method introduces causal counterfactual debiasing theory with a calibration network to mitigate LLM biases while improving generalization. Extensive experimental results demonstrate that the proposed method achieves superior performance over the state-of-the-art baselines.
KW - Causal debiasing
KW - Chain-of-Thought
KW - Data augmentation
KW - Zero-shot stance detection
UR - https://www.scopus.com/pages/publications/105022695405
U2 - 10.1007/978-981-95-4091-4_14
DO - 10.1007/978-981-95-4091-4_14
M3 - Conference contribution
AN - SCOPUS:105022695405
SN - 9789819540907
T3 - Communications in Computer and Information Science
SP - 196
EP - 212
BT - Neural Information Processing - 32nd International Conference, ICONIP 2025, 2025, Proceedings
A2 - Taniguchi, Tadahiro
A2 - Leung, Chi Sing Andrew
A2 - Kozuno, Tadashi
A2 - Yoshimoto, Junichiro
A2 - Mahmud, Mufti
A2 - Doborjeh, Maryam
A2 - Doya, Kenji
PB - Springer Science and Business Media Deutschland GmbH
T2 - 32nd International Conference on Neural Information Processing, ICONIP 2025
Y2 - 20 November 2025 through 24 November 2025
ER -