<XML><RECORDS><RECORD><REFERENCE_TYPE>3</REFERENCE_TYPE><REFNUM>8306</REFNUM><AUTHORS><AUTHOR>Ren,R.</AUTHOR><AUTHOR>Jose,J.M.</AUTHOR></AUTHORS><YEAR>2006</YEAR><TITLE>Attention Guided Football Video Content Recommendation on Mobile Device</TITLE><PLACE_PUBLISHED>MobileMedia 2006, Italy</PLACE_PUBLISHED><PUBLISHER>ACM Press</PUBLISHER><LABEL>Ren:2006:8306</LABEL><KEYWORDS><KEYWORD>Attention feature</KEYWORD></KEYWORDS<ABSTRACT>Live football video is the major content genre in 3G mobile service. In this paper, we introduce a realtime general highlight detection algorithm based on attention analysis. It combines attention-related media modalities into role-based attention curves, namely video director, spectator and commentator, to track viewers' feeling against game content from media data. A series of linear temporal predictors are generated from video data directly and employed to allocate strong attention changes, which are marked as scroll-back endpoints for mobile video skim. The advantages of our algorithm are that it avoids semantic uncertainty of game highlights and requires little training. We evaluated our approach using a test bed with five full games in FIFA World Cup 2002 and European League 2006 from different content suppliers, i.e. BBC and ITV to prove the robustness.</ABSTRACT></RECORD></RECORDS></XML>