Conditional Information Bottleneck for Multimodal Fusion: Overcoming Shortcut Learning in Sarcasm Detection

Yihua Wang; Qi Jia; Cong Xu; Feiyu Chen; Yuhan Liu; Haotian Zhang; Liang Jin; Lu Liu

AAAI 2026technical0 citations

Conditional Information Bottleneck for Multimodal Fusion: Overcoming Shortcut Learning in Sarcasm Detection

Yihua Wang, Qi Jia, Cong Xu, Feiyu Chen, Yuhan Liu, Haotian Zhang, Liang Jin, Lu Liu

Abstract

Multimodal sarcasm detection is a complex task that requires distinguishing subtle complementary signals across modalities while filtering out irrelevant information. Many advanced methods rely on learning shortcuts from datasets rather than extracting intended sarcasm-related features. However, our experiments show that shortcut learning impairs the model

BibTeX

@inproceedings{aaai2026_conditionalinfor,
  title = {Conditional Information Bottleneck for Multimodal Fusion: Overcoming Shortcut Learning in Sarcasm Detection},
  author = {Yihua Wang and Qi Jia and Cong Xu and Feiyu Chen and Yuhan Liu and Haotian Zhang and Liang Jin and Lu Liu and Zhichun Wang},
  booktitle = {AAAI 2026},
  year = {2026}
}