Building multimodal AI that can read affect, model minds, and recover what people mean beyond what they say. CogAFFC is motivated by a core gap: multimodal systems can often recognize visible emotion, yet still fail to explain why it emerges, what mental state causes it, and what hidden social meaning is being conveyed. It connects multimodal perception, Theory-of-Mind reasoning, implicit social context, and empathetic intelligence into one coherent agenda.
Flagship Research
Start from the observable world: facial expression, speech, language, video context, and other cross-modal cues that reveal emotional state.
Move from what is seen to what is mentally simulated: beliefs, intentions, causes, and intermediate cognitive states that explain emotion.
Go beyond explicit emotion to what people actually imply under social constraints, politeness, strategic expression, irony, and hidden stance.
THOR-ISA studies implicit sentiment analysis, where opinion cues appear in obscure and indirect forms. The work introduces a three-hop Chain-of-Thought prompting framework that progressively infers latent aspect, opinion, and final sentiment polarity.
HitEmotion reframes multimodal emotion understanding as a cognition-sensitive task. It introduces a Theory-of-Mind-grounded benchmark and a reasoning pipeline that explicitly tracks mental states instead of relying on shallow post-hoc rationales.
MoCA formalizes the problem of recovering what people truly feel, intend, or imply when their expression is indirect. The work expands the thread from explicit emotion understanding to latent social meaning under multimodal social context.
Community
The workshop anchors the community around cognition-oriented multimodal affective and empathetic computing. It frames the area around context, causal reasoning, emotion understanding, and the next generation of affect-aware multimodal models.