Interpreting and Enhancing Emotional Circuits in Large Vision-Language Models via Cross-Modal Information Flow

• URL: http://arxiv.org/abs/2605.21980v1
• Date: Thu, 21 May 2026 04:20:12 GMT
• Title: Interpreting and Enhancing Emotional Circuits in Large Vision-Language Models via Cross-Modal Information Flow
• Abstract summary: Internal mechanisms governing how LVLMs translate abstract visual stimuli into coherent emotional narratives remain largely unexplored.<n>We introduce a steering-vector-based causal attribution framework tailored for descriptive emotional reasoning.<n>We regulate the emotional information routing to strengthen attention flow and amplify the semantic activation to consolidate expression.
• Score: 12.059256609530294
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large Vision-Language Models (LVLMs) represent a significant leap towards empathetic agents, demonstrating remarkable capabilities in emotion understanding. However, the internal mechanisms governing how LVLMs translate abstract visual stimuli into coherent emotional narratives remain largely unexplored, primarily due to the scarcity of visual counterfactuals and the diffuse nature of emotional expression. In this paper, we bridge this gap by introducing a steering-vector-based causal attribution framework tailored for descriptive emotional reasoning. To this end, we construct a specialized dataset to demystify the emotional circuits underlying the three-stage ``Adapt-Aggregate-Execute'' mechanism. Crucially, we discover a functional decoupling: visual emotional cues are aggregated in middle layers via sentiment-specific attention heads, but are subsequently translated into narrative generation in deep layers through emotion-general pathways. Guided by these insights, we regulate the emotional information routing to strengthen attention flow and amplify the semantic activation to consolidate expression. Extensive experiments on the comprehensive MER-UniBench demonstrate that our methods significantly improve performance via inference-time intervention, effectively mitigating emotional hallucinations and corroborating the causal fidelity of the discovered circuits.

Related papers

• EII-SCL: Harnessing Emotional Inertia for Multimodal Emotion Recognition in Conversation [51.04910642896281]
  Multimodal emotion recognition in conversation (MERC) achieves accurate predictions by integrating multimodal and contextual information in dialogues.<n>Current MERC approaches overlook the impact of contextual emotional inertia in emotion shift, leading to sub-optimal performance.<n>We propose a novel Emotional Inertia-Informed Supervised Contrastive Learning module (EII-SCL) that informs the contrastive objective by constructing inertia-affected samples within temporal windows.
  arXiv  Detail & Related papers  (2026-07-19T18:09:54Z)
• From Syntax to Emotion: A Mechanistic Analysis of Emotion Inference in LLMs [1.4692712025142225]
  We investigate the internal mechanisms of emotion recognition in large language models (LLMs)<n>By analyzing sparse feature activations across layers, we identify a consistent three-phase information flow.<n>We show that emotion representations comprise both shared features across emotions and emotion-specific features.
  arXiv  Detail & Related papers  (2026-04-28T17:03:16Z)
• EMO-R3: Reflective Reinforcement Learning for Emotional Reasoning in Multimodal Large Language Models [62.3977734456669]
  We propose Reflective Reinforcement Learning for Emotional Reasoning (EMO-R3), a framework designed to enhance the emotional reasoning ability of Multimodal Large Language Models (MLLMs)<n>We introduce Structured Emotional Thinking to guide the model to perform step-by-step emotional reasoning in a structured and interpretable manner, and design a Reflective Emotional Reward that enables the model to re-evaluate its reasoning based on visual-text consistency and emotional coherence.<n>EMO-R3 significantly improves both the interpretability and emotional intelligence of MLLMs, achieving superior performance across multiple visual emotional understanding benchmarks.
  arXiv  Detail & Related papers  (2026-02-27T08:42:52Z)
• Emotion-Coherent Reasoning for Multimodal LLMs via Emotional Rationale Verifier [53.55996102181836]
  We propose the Emotional Rationale Verifier (ERV) and an Explanation Reward.<n>Our method guides the model to produce reasoning that is explicitly consistent with the target emotion.<n>We show that our approach not only enhances alignment between explanation and prediction but also empowers MLLMs to deliver emotionally coherent, trustworthy interactions.
  arXiv  Detail & Related papers  (2025-10-27T16:40:17Z)
• Anatomy of a Feeling: Narrating Embodied Emotions via Large Vision-Language Models [1.8349570933241344]
  We propose a framework to generate Embodied LVLM Emotion Narratives (ELENA)<n>These are well-defined, multi-layered text outputs that focus on the salient body parts involved in emotional reactions.<n>We observe that our employed framework can effectively recognize embodied emotions in face-masked images, outperforming baselines without any fine-tuning.
  arXiv  Detail & Related papers  (2025-09-23T21:34:57Z)
• From Coarse to Nuanced: Cross-Modal Alignment of Fine-Grained Linguistic Cues and Visual Salient Regions for Dynamic Emotion Recognition [7.362433184546492]
  Dynamic Facial Expression Recognition aims to identify human emotions from temporally evolving facial movements.<n>Our method integrates dynamic motion modeling, semantic text refinement, and token-level cross-modal alignment to facilitate the precise localization of emotionally salient features.
  arXiv  Detail & Related papers  (2025-07-16T04:15:06Z)
• Emotion-Qwen: A Unified Framework for Emotion and Vision Understanding [26.36195886824082]
  Emotion-Qwen is a unified multimodal framework designed to simultaneously enable robust emotion understanding and preserve general reasoning capabilities.<n>We develop the Video Emotion Reasoning dataset, a large-scale bilingual resource containing over 40K video clips annotated with detailed context-aware emotional descriptions.
  arXiv  Detail & Related papers  (2025-05-10T16:15:26Z)
• Disentangle Identity, Cooperate Emotion: Correlation-Aware Emotional Talking Portrait Generation [63.94836524433559]
  DICE-Talk is a framework for disentangling identity with emotion and cooperating emotions with similar characteristics.<n>We develop a disentangled emotion embedder that jointly models audio-visual emotional cues through cross-modal attention.<n>Second, we introduce a correlation-enhanced emotion conditioning module with learnable Emotion Banks.<n>Third, we design an emotion discrimination objective that enforces affective consistency during the diffusion process.
  arXiv  Detail & Related papers  (2025-04-25T05:28:21Z)
• SOLVER: Scene-Object Interrelated Visual Emotion Reasoning Network [83.27291945217424]
  We propose a novel Scene-Object interreLated Visual Emotion Reasoning network (SOLVER) to predict emotions from images. To mine the emotional relationships between distinct objects, we first build up an Emotion Graph based on semantic concepts and visual features. We also design a Scene-Object Fusion Module to integrate scenes and objects, which exploits scene features to guide the fusion process of object features with the proposed scene-based attention mechanism.
  arXiv  Detail & Related papers  (2021-10-24T02:41:41Z)
• Stimuli-Aware Visual Emotion Analysis [75.68305830514007]
  We propose a stimuli-aware visual emotion analysis (VEA) method consisting of three stages, namely stimuli selection, feature extraction and emotion prediction. To the best of our knowledge, it is the first time to introduce stimuli selection process into VEA in an end-to-end network. Experiments demonstrate that the proposed method consistently outperforms the state-of-the-art approaches on four public visual emotion datasets.
  arXiv  Detail & Related papers  (2021-09-04T08:14:52Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.