Enriching videos with light semantics

This paper describes an ongoing prototypical framework to annotate and retrieve web videos with light semantics. The proposed framework reuses many existing vocabularies along with a video model. The knowledge is captured from three different information spaces (media content, context, document). We also describe ways to extract the semantic content descriptions from the existing user-generated content using multiple approaches of linguistic processing and Named Entity Recognition, which are later identified with DBpedia resources to establish meanings for the tags. Finally, the implemented prototype is described with multiple search interfaces and retrieval processes. Evaluation on semantic enrichment shows a considerable (50% of videos) improvement in content description.