The value of stories for speech-based video search

Anecdotal evidence suggests that story-level information is important for the speech component of video retrieval. In this paper we perform a systematic examination of the combination of shot-level and story-level speech, using a document expansion approach. We isolate speech from other retrieval features, and evaluate on the 2003–2006 TRECVID test sets with a set of 94 natural language queries. Our main finding is that that the use of story information significantly improves retrieval performance compared to shotbased search, increasing overall mean average precision by over 65%. Categories and Subject Descriptors H.3 [Information Storage and Retrieval]: H.3.1 Content Analysis and Indexing; H.3 .3 Information Search and Retrieval; H.3.4 Systems and Software; H.4 [Information Systems Applications]: H.4.2 Types of Systems; H.4.m Miscellaneo us General Terms Algorithms, Measurement, Performance, Experimentation Keywords Story information, video news retrieval, temporal mismatch

• Type: Conference
• Year: 2007
• Where: CIVR
• Authors: Bouke Huurnink, Maarten de Rijke
• Keywords: CIVR 2007, Retrieval Features, Retrieval Performance, Video Retrieval