Video search : Blinkx Blinkx
Jason Pontin runs a glowing article on Blinkx and their video search technology. Speech recognition for auto-tagging, in a way that scales, is the very short pitch for Blinkx. They claim 7 million videos, and the search experience is certainly pretty fast.
Video Search is a major headache but particularly for UGC video companies such as DailyMotion, where users just don’t tag well. In professionally produced libraries, I can imagine that meta-data is fairly decent, but in my world tag spamming, misspellings and laziness all conspire to provide very low quality explicit data.
I missed my chance of meeting the founders last Friday to ask all the burning questions I have about the tech behind the emerging brand that Blinkx has been building. Here is a beet.tv video as a proxy.
Ton console myself, I tried the nice wall feature that Blinkx have been using as their viral feature. Do try it out on their site! After a few frustrating attempts, I just used "Rodrigo" as a search term, trying to check how much would be Rodrigo-Sepulveda related. Fairly unique name after all, and vpod.tv did a deal with them. I know of course this is not a narrow enough search, but I like the random "stumble upon" feel you get (and quite a few of the videos were indeed captured by Rodrigo S., including an MRY show :-)). I tried Paris too as a term but that was loaded with porn (courtesy of Paris Hilton tag spam) and not really publishable, which highlights one clear issue for the new search hopefuls. Blinkx from the user standpoint: great wow factor, but you will need to be very smart about your search terms for real relevancy.