TCR: Short Video Title Generation and Cover Selection with Attention Refinement

With the widespread popularity of user-generated short videos, it becomes increasingly challenging for content creators to promote their potential viewers. Automatically generating appealing titles and covers videos can help grab viewers’ attention. Existing studies on video captioning mostly focus factual descriptions actions, which do not conform intended catching viewer Furthermore, research cover selection based multimodal information is sparse. These problems motivate need tailored methods specifically support joint task title generation (TG-CS) as well demand creating corresponding datasets studies. In this paper, we first collect present a real-world dataset named Short Video Title Generation (SVTG) that contains with covers. We then propose Cover attention Refinement (TCR) method TG-CS. The refinement procedure progressively selects high-quality samples highly relevant frames text tokens within each sample refine model training. Extensive experiments show our TCR superior various existing in able select better noisy videos.