Generic descriptions for movie document: An experimental study

Abstract

The description of movie documents has become a necessary condition to explore the content. For this purpose, the extraction of generic descriptions represents a pertinent solution for an interrogation process. In this context, two types of descriptions are used, an external and a content description. In this regard, the title, the genre and the keyword descriptions represent the description that can be identified from the content. Consequently, we propose in this paper a framework not only to extract these descriptions, but also to define a relationship between the movies through these descriptions. In fact, to extract the external description, we use the Dublin Core metadata standard while to extract the content description, we propose a method to identify the genre and the keyword. The objective behind this method is to extract the semantic knowledge from the content. Indeed, two types of techniques are introduced, namely, statistical and semantic. This principle is based on the use of the combination between textual modality through the synopsis, the script of the film and the visual modality through the use of the superposed text on the image. The experimental results confirmed a promising performance with two databases, namely, 'MovieLens dataset' and 'MEG database' through the following validation techniques: pertinence, recall, precision and F-measure.