Enrichment and Ranking of the YouTube Tag Space and Integration with the Linked Data Cloud

  • Authors:

  • Smitashree Choudhury;John G. Breslin;Alexandre Passant

  • Affiliations:

  • DERI, National University of Ireland, Galway, Ireland;DERI, National University of Ireland, Galway, Ireland and School of Engineering and Informatics, National University of Ireland, Galway, Ireland;DERI, National University of Ireland, Galway, Ireland

  • Venue:
  • ISWC '09 Proceedings of the 8th International Semantic Web Conference
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

The increase of personal digital cameras with video functionality and video-enabled camera phones has increased the amount of user-generated videos on the Web. People are spending more and more time viewing online videos as a major source of entertainment and "infotainment". Social websites allow users to assign shared free-form tags to user-generated multimedia resources, thus generating annotations for objects with a minimum amount of effort. Tagging allows communities to organise their multimedia items into browseable sets, but these tags may be poorly chosen and related tags may be omitted. Current techniques to retrieve, integrate and present this media to users are deficient and could do with improvement. In this paper, we describe a framework for semantic enrichment, ranking and integration of web video tags using Semantic Web technologies. Semantic enrichment of folksonomies can bridge the gap between the uncontrolled and flat structures typically found in user-generated content and structures provided by the Semantic Web. The enhancement of tag spaces with semantics has been accomplished through two major tasks: (1) a tag space expansion and ranking step; and (2) through concept matching and integration with the Linked Data cloud. We have explored social, temporal and spatial contexts to enrich and extend the existing tag space. The resulting semantic tag space is modelled via a local graph based on co-occurrence distances for ranking. A ranked tag list is mapped and integrated with the Linked Data cloud through the DBpedia resource repository. Multi-dimensional context filtering for tag expansion means that tag ranking is much easier and it provides less ambiguous tag to concept matching.