EmoMM: Benchmarking and Steering MLLM for Multimodal Emotion Recognition under Conflict and Missingness
This work addresses the underexplored issue of MLLM decision-making under modality conflict and missingness for emotion recognition, offering a benchmark and a practical solution to improve reliability.
The paper introduces EmoMM, a benchmark for multimodal emotion recognition under modality conflict and missingness, and proposes CHASE, a lightweight attention steering method that mitigates decision bias in MLLMs, consistently improving performance across settings.
Multimodal Emotion Recognition (MER) is critical for interpreting real-world interactions. While Multimodal Large Language Models (MLLM) have shown promise in MER, their internal decision-making mechanisms under modality conflict and missingness remain largely underexplored. In this paper, to systematically investigate these behaviors, we introduce EmoMM, a comprehensive benchmark featuring modality-aligned, conflict, and missing subsets. Through extensive evaluation, we uncover a Video Contribution Collapse (VCC) phenomenon, where MLLM marginalize video evidence due to high token redundancy and modality preferences. To address this, we propose Conflict-aware Head-level Attention Steering (CHASE), a lightweight mechanism that detects modality conflicts and performs inference-time attention steering, effectively mitigating decision bias without retraining the backbone. Experimental results demonstrate that CHASE consistently improves performance across various settings, significantly enhancing the reliability of MLLM in complex affective scenarios.