Vikas Kumar

LLM-generated summaries can vary significantly in trustworthiness, with some models exhibiting up to 40% instability in their outputs across different document types.

Language is the dominant channel for detecting ambivalence and hesitancy, but leveraging ASR-erased time features can significantly boost recognition accuracy.

DReX offers a unified multimodal recommendation framework that not only outperforms existing methods but also provides interpretable user and item profiles by leveraging interaction-level features.

Forget isolated user histories – GrIT leverages dynamic group affiliations to predict your next purchase with surprising accuracy.