Enriching media fragments with named entities for video classification