Telegenic Data Analysis for Compact Description and Genre Classification based on Re-ranking

Extractive summarization aims to shorten textual information to create short summary by selecting sentences enriched with information. Ranking of sentences based on the previously selected sentences can assist in generating summaries for data analysis. In proposed research paper, a policy gradient-based reinforcement learning approach has been discussed to re-rank sentences by selecting high score sentence. Experiments have been conducted on TREC video dataset of rushes video and ROUGE evaluation metric clearly depicts the better performance of the reinforcement model with Rouge score of 74.39%. After generating compact telegenic descriptions, videos are classified into seven categories including: news, action, outdoor/indoor scene setting, close-up, grouping, traffic and meeting using Random Forest classifier with accuracy of 95.73%.