A Multitask Causality-Inspired Feature Enhancement Method for Stance Detection

Jian Xu, Bo Liu, Yanshan Xiao

Abstract: The existing stance detection methods have several limitations. (1) They utilize additional information such as sentiment or linguistic features to construct multitask frameworks for achieving performance bottleneck breakthroughs and only use the last encoder layers of their language models for semantic encoding. (2) They may establish spurious correlations between input and labels based on statistical dependence, causing text-target representations to contain more biased stance-unrelated noncausal features, which act as shortcuts for stance prediction without considering causal mechanisms, leading to performance bottlenecks. In this paper, we propose a novel multitask causality-inspired feature enhancement (MTCIFE) method for stance detection, achieving a performance breakthrough without additional information. MTCIFE introduces two stance detection tasks: an auxiliary stance detection task (ASDT) and a main stance detection task (MSDT), constructing the relation between text and a target in different ways by using different transformer encoder layers of bidirectional encoder representations from transformers (BERT); this augments the diversity of the obtained representations and makes them learn from each other. Then, we decouple stance-related causal features from stance-unrelated noncausal features and encourage their independence in both tasks. Considering the underlying causal mechanisms, we propose a causality-inspired feature enhancement (CIFE) module for implementing causal learning and intervention at the feature level. By integrating the CIFE module into both tasks via multitask learning, we aim to decouple the learning of causal and noncausal features, improving causal feature acquisition and mitigating the confounding effect of noncausal features. Extensive experiments demonstrate the outstanding performance of our model over other state-of-the-art approaches.

External IDs:doi:10.1109/taslpro.2025.3536168