COSI-Lab: New Dataset Captures Social Intentions in Scientific Workshops

COSI-Lab has been launched by researchers as a dataset featuring multimodal and multi-sensor data from an interdisciplinary scientific workshop attended by 32 scholars at an international conference. This dataset records authentic social interactions in a loosely structured environment, comprising two 30-minute mingling sessions with genuine professional and social implications for the participants. The investigation centers on the Apparent Intent Inference (AII) challenge, as assessed by external observers, framing intentions as separate from observable future results. The authors suggest that intelligent systems of the future could enhance subjective perception management by treating multiplicity as an explainable reasoning process rather than mere label noise. They propose a new annotation method for AII that incorporates the perceiver's interpretative biases and includes both quantitative and qualitative analyses of intent narratives regarding diversity and grounding. This dataset aims to facilitate research in social intention modeling and multimodal interaction analysis. The paper can be found on arXiv with the identifier 2607.28649.