Framework for context-based film recommendation system

This paper proposes a novel framework for a context-based film recommendation system based on shot genre classification of MPEG compressed films. This method analyzes low level audio-visual features extracted on compressed domain, to then classify MPEG coded films into predefined genres at the shot level to allow efficient indexing. Moreover, the distribution of the resulting shot genres can be used to represent film context, which is in turn applied to a film recommendation system. According to reasonable classification results in some experiments using MPEG-1 coded films, the proposed film recommendation system enables to recommend subjectively appropriate films to users without large dataset of film ratings and user preferences on film genres, which has not been achieved by conventional methods such as content-based approaches.