OntoVidRec: A Staged Knowledge Aggregation Scheme for Annotations-Based Video Retrieval Using Ontology Matching

Abstract

The amounts of multimedia data on the Web have increased in recent times and there are very few approaches for video recommendations. There is a need for a semantically driven video recommendation system as the current structure of the Web is progressing into a rational Semantic Web. The increase in the number of videos has also enhanced the annotations and the metadata associated with the video recommendation system. Due to the luxury of the availability of an increase in the amounts of metadata on the Web specifically for the video contents, an annotations-based video retrieval scheme by aggregating the meta information of the video is a mandated requirement as it can reduce the computational complexity in video recommendation than the traditional content-based video recommendation systems. In this paper, an OntoVidRec framework which aggregated video-related meta information from the query, user profiles, and the dataset has been proposed. The approach builds formal ontologies from the individual sources and encompasses a strategic model for Ontology Matching to yield the most appropriate Query relevant Ontological Entities that are semantically matched with the video annotations and are recommended based on the semantic similarity and the Kullback-Leibler divergence measure. An overall F-Measure of 95.37% has been achieved by the OntoVidRec which is the best in class performance for such systems.