K consecutive clips that cover the content material of the whole synopsis paragraph. Then we develop a coarse-to-high-quality procedure to effectively align each paragraph to its corresponding phase. 2) Each synopsis paragraph is dispatched to 3333 annotators. Making use of our validation set, we observe that larger ranks within the approximation do not give better outcomes and by trying lower values, we reach at the outcomes seen in Fig. 2. As could be seen from Fig. 2, prediction results improve until rank 3333 after which lower.