CIL2007: Engines for multimedia search
Came in about 15 minutes late (tried another session but it was packed)
Ran Hock
He’s talking now about image search and how the available fields are important in this. Some features are common across.
Flickr has some unique fields
Interestingness (possibly, he says, using implicit relevance feedback)
Camera (he doesn’t see the utility – but for photographers who are comparing models, could be useful)
There is an advanced search, but it’s tricky to find
AudioSame characters, plus AOL (surprisingly enough), Internet Archive
Other considerations – cached copy at the search engine?
Yahoo – limit to podcasts or other audio, expand to get more options like format, duration.
Exalead – regular search and then you limit it – on the right hand side you get facets to narrow on
AOL – includes search formerly known as singing fish.
Internet Archive – 133,000 (compare to 50M in Yahoo) including live concert recordings (Grateful Dead as a category :) )
Audio – Podcast
Sometimes a lot better metadata – up to full transcripts available to search (either machine or human produced), enclosures in XML so there’s a lot of associated text
Podscope – automatic speech recognition (ASR) – last time when Greg showed this it came up with some funny results.
R.H. doesn’t find the search in iTunes helpful.
Audio – Music
Yahoo music has comparison shopping when you find songs you want to buy
He doesn’t mention Pandora (which has the way coolness factor in my mind)
Video(seems to me very different from last year with the explosion of YouTube)
He showed us a really, really cool Panama Canal video – he points out how much more info they give on the right hand side – facets to pivot on and do new searches on.
“legitimate” clips to purchase
Video – TV
TVeyes (fee based – has this changed?), transcripts, $500+/mo, good for companies who need to monitor news
Shadow TV
Blinkx (not as current, big database lots of sources, advanced search)
Labels: cil2007