Movie Summarization based on Indonesian Subtitles with Restricted Boltzmann Machine

A summary of a movie provides a concise description of the movie both concretely and abstractly. Automatic movie summarization can be applied to get a movie summary. In this study, we propose a movie summarization method on Indonesian subtitles with five stages, namely: preprocessing, feature extraction, feature enhancement, sentence ranking, and summary generation. The preprocessing consist of document segmentation, paragraph segmentation, case folding, normalization, stopword removal, and part-of-speech tagging. Twelve intrinsic features, namely: number of thematic words, sentence position, sentence length, sentence position relative to paragraph, number of proper noun, number of numerals, number of verbs, number of noun, term frequency-inverse sentence frequency, sentence to centroid similarity, bi-gram key phrase list, and tri-gram key phrase list are extracted and enhanced with Restricted Boltzmann Machine (RBM) to improve the quality of the summary. Results indicate that the summary produced is understandable and has the name of the main character in the provided text. Compare to movie summarization without RBM, RBM improves the average quality of the summary of 6% (F1-score). Questionnaire results also show that the movie summarization developed in this study produces a summary that is suitable for the provided subtitles.