AOL Buys Video Search Firm

Eric Newman writes "TheStreet.com is reporting that America Online has purchased Truveo.com. From the article: 'Truveo has a proprietary technology called visual crawling that lets it automatically discover video files on Web pages, enabling customers to see updated information on news, sports and entertainment. The acquisition, which closed Dec. 21, was AOL's fifth last year. News of the deal wasn't released until Tuesday. Terms were not disclosed.' Note that the deal closed the same week that Google bought a 5% stake in AOL, in part to collaborate on video technology."

Comments

the true meaning

[User 956]

'Truveo has a proprietary technology called visual crawling that lets it automatically discover video files on Web pages, enabling customers to see updated information

And by "updated information", of course, they mean "porn".

Didn't Google

[dbucowboy]

already implement an effective video search technology?

How?

[tacokill]

How, excatly, does this thing work? I am not an expert in search technologies but one thing that jumps out at me is this:

How do you index videos and put context around them?

Does it parse the language that is being spoken? Does it read the subtitles? For example, if I snip a 1 minute story on the G-7 summit from CNN, how do you know what the story is about if I don't tell you? To my knowledge, there is no sophisticated technology solution for this aside from reading the subtitles and indexing that.

I've thought about this alot. Everyone and their dog seems to be coming out with a video search engine of somekind and not a single implementation has explained how they are going to do the indexing.

I suppose they could take the Yahoo approach and view/sort each video that is submitted. But that is not a realistic long term solution, IMHO.

Re:How?

[Otter]

Reading Truveo's site, it looks like a) it gets all its metadata from the context in which it's found, not from examining the video itself and b) their real accomplishment is being able to fish through the different layers behind which most news sites put their video. It doesn't sound super-imprressive to me, but AOL seems to disagree.

As for your question -- presumably you could pipe the audio portion to a speech-to-text tool and parse that, no?

Re:How?

[Big_Al_B]

The simple technical solution is to embed meta-data in tags.

Some of that can be automatically populated, such as creation date, length and file type. Some has to be manually added, such as title, rating, or genre.

Re:How?

[MoonFog]

So it basically does for videos what Google image search does for images if I've understood this correctly? Not that there's anything wrong with that, it works alright with Google images so far.

Re:How?

[OakDragon]

A quick search turned up two abstracts for video search algorithms:

A Fast Multi-Resolution Block Matching Algorithm for Multiple-Frame Motion Estimation

Efficient Video Similarity Measurement and Searc (probably grad students here)

I felt my brain being damaged while I looked them over, but they appear to employ something similar to image matching with the added component of movement. It looks like if they are implemented as desired, you could find video similar to a reference piece. This is not useful for searching based on a text query, however. But, you could build an index that matched words to a reference library of video clips, then search for matches to your reference clip.

Of course, all the heavy crunching would be used to build a lean, fast search index, hopefully.

Re:How?

[Otter]

So it basically does for videos what Google image search does for images if I've understood this correctly?

If I understood the Truveo site correctly -- yeah, it's similar to Google image search except with a supposedly better crawler.

I tried visual crawling once

[digitaldc]

but I only saw a slow-moving floor.

But seriously folks, the search engine works rather well. Its interesting to note that the ads on truveo are by google, and http://video.google.com/ is another viable alternative.

Even though we are just getting started, we have already indexed an extensive collection of web video that you will not find in any other search engine.
Google will soon take care of that.

yawn... wake me up when its something new

[skiingyac]

truveo.com is basically google image but for video... Take google's image search crawler logic, replace jpg with mpg (or rather wmv or flash...) and voila. explain why truveo.com is so awesome?

You've got

[OSS_ilation]

Video! Now, please wait while new content is loaded. Loading...

AOL has a New Video Portal

[microbrewer]

Check out AOL Hi-Q Video http://hiqvideo.aol.com/

Its video delivered by Kontiki's p2p grid technology .

Wonder if Google will end up delivering content in this Manner .

Wake Up

[billsoman]

It is nothing like Google's strategy, which (like Yahoo's) is inherently limited and hyper-expensive, depending as it does on doing deals to arrange feeds from content providers, who can decide what to share and who will eventually and inevitably extract their pound of flesh. Truveo's technology does not depend on feeds, nor metadata (notoriously bad). Their servers instantiate and visually crawl the content, including dynamic content, and extract information from the content itself and its context - including subtitles where present and even speech-to-text if necessary. Pulling that off at scale is super hard, and Truveo is doing it. Very cool, ground-breaking, and apparently patent-pending stuff.

RTFA and TTFP before posting, please.