Youku Technology

May 18, 2020 · Artificial Intelligence

How Feature-Induced Manifold Disambiguation Improves Video Tagging in Multi-View Learning

The paper "Feature‑Induced Manifold Disambiguation for Multi‑view Partial Multi‑label Learning" accepted at KDD 2020 introduces the MVPML framework and the FIMAN method, which leverage heterogeneous multimodal features to correct and supplement video tags, thereby boosting distribution efficiency in Alibaba Entertainment’s platforms.