Kansei-keyword extraction from Japanese film scenario using sensitivity information

A method of extracting kansei-keyword from Japanese film scenarios based on their sensitivity information is proposed. The kansei-keyword extracted from each scene in a film is useful as the index to retrieve scenes in the film. In addition to the relations between kansei-keywords (usually paired with the antonymous keyword) and sensitivity expressions, which has been used in former studies, the relations among sensitivity expressions within the same sentence are also considered to extract kansei-keywords from scenarios. The 20 pairs of kansei-keywords are selected based on the questionnaire to 50 people, who frequently watch the movies, while the relations between selected kansei-keywords and 1171 sensitivity expressions are determined based on another questionnaire to 27 people. The kansei-keywords of a scene are determined based on the impressive information of all sentences in it, which are calculated from the relation between sensitivity expressions. Experiments are done with Japanese film scenario, "Shall we dance?", and it is confirmed that both of the recall rate and the precision are 70%, which is sufficient for the applications such as a scene retrieval systems.