SEAL: Semantic Attention Learning for Long Video Representation | Read Paper on Bytez