Complex　named　entities　can　refer　to　specific　objects,　literal　obvious　characteristics,　which　are　closely　related　entities　in　our　daily　life,　study　and　work.　Named　entity　recognition　complex　cannot　form　only　an　important　data　resources　on　the　Internet　to　provide　a　basis　for　information　extraction,　but　also　leverage　search　engines　to　help　to　understand　the　user＇s　query　intent,　which　gives　targeted,　integrated　search　results.　This　paper　proposes　a　WEB　video　proposes　a　web　video　oriented　model　to　extract　complex　named　entities.　The　text　around　the　video-based　information　(mainly　the　labels　and　categories　of　video　information),　in　accordance　with　the　characteristics　of　various　types　of　word-class　distribution　and　the　between-class　distribution,　and　the　complex　characteristics　of　named　entities　and　the　relationship　between　co-occurrence　of　words,　extracted　from　the　various　categories　of　named　entity.　In　our　method,　a　small　amount　of　word　is　marked　for　each　category.　The　corpus　does　not　require　a　large　number　of　marked　and　prolonged　training　process,　which　greatly　increases　the　complexity　of　the　named　entity　extraction　accuracy　and　efficiency.　©　2010　IEEE.