A2Gen: Action-Aware Generative Model for Short Video Recommendation

A new paper on arXiv proposes A2Gen, an action-aware generative sequence network for short video recommendation. Traditional models treat videos as holistic entities, missing nuanced user preferences across diverse segments. By analyzing user action timing as a temporal process, A2Gen refines actions along the temporal dimension and chains them for unified prediction. The model includes a Context-aware Attention Module. The paper is authored by researchers and published on arXiv under ID 2604.25834.