Thursday, October 11, 2007

Opportunity 2: Video Search and Discovery


Video Search and Discovery are two very important, and closely related, functions: Discovery is when you want to find something to watch, and search is when you want to find something specific to watch. When I was writing the report on Internet Video for MRG, I wanted to illustrate the problems with video search, and I decided to do a very simple search, on the word "Panama." There are lots of things containing the word Panama: Panama the country, the Panama Canal, Panama hats, etc., but I was interested in the country and things related to the country.

I started with a text search using the Google engine, and sure enough, the first page of the search was exclusively related to the country and things in the country (including the Canal.) I then did the same search with a variety of video search engines: YouTube, Dabble, Truveo and EveryZing. (After the report went to press, I did the same test with blinkx, and got similar results.)

YouTube and Dabble gave the best results--everything was either related to the country Panama or had been produced in the country. Blinkx returned videos about Panama, the Canal, Panama City, Florida, and three versions of the Van Halen song "Panama." Truveo came up with a seemingly random list of videos with the word "Panama" somewhere in the title, description or tags. And EveryZing was the worst--it relies on speech to text recognition, and it completely misrecognized the speech in the videos it returned. Not one of the videos had anything to do with Panama.

Good video search relies on two components:

  1. The metadata (title, description and tags) associated with each video, and
  2. What the search engine does with that metadata.

Search engines like Google and Yahoo! have it relatively easy--they can use all the text in a page in order to index the page's content. The more text they've got, the more accurately they can index the page. However, to automatically index videos, you have to rely on the amount and accuracy of the metadata, not the content of the videos themselves. In my experience, search engines like EveryZing that reply primarily or exclusively on speech-to-text recognition simply aren't "ready for prime time."

YouTube gives good (not great) results, because it processes whatever metadata is available with Google's search engine. Dabble gets equally good (some would say better) results because individuals review and tag all the videos, and better metadata means better results.

In my opinion, however, no one has "cracked the code" for video search. This remains an area of tremendous opportunity for at least three companies (the companies that will eventually be bought by Google, Yahoo and Microsoft.)

Discovery is the other side of the coin. It's what you do when you're "channel surfing," or when you're scanning through an Electronic Program Guide (EPG) to find something to watch. EPGs have been around for many years, and they've worked very well. However, as the number of channels increases and Video-on-Demand (VOD) services become more popular, it gets harder and harder to find something to watch with an EPG. Including HD and music channels, many cable, satellite and IPTV systems now have channels into the 900's. How is a viewer going to browse through so many choices?

Companies are working on solutions to make video discovery easier and more intuitive. Hillcrest Labs in the U.S. and Ruwido in Austria have developed custom remote controls and software designed to make it simpler to navigate large libraries of broadcast and VOD content.

At the recent IBC conference in Amsterdam, Orca Interactive of Israel was claiming that "the EPG is dead." They and other companies have licensed collaborative filtering software called ContentWise from Neptuny of Italy. ContentWise both enables viewers to specify the categories and keywords that they're interested in and monitors actual behavior to determine viewers' interests. Then, programming that meets the viewer's interests as determined by ContentWise is offered on-screen, using a "cross-hair" display instead of the typical EPG table of channels and programs.

At IBC, we were assured that ContentWise won't be susceptible to "Amazon disease". If you're an Amazon customer, you know what I'm talking about--you purchased a book for a friend as a gift years ago, but Amazon keep trying to sell you similar books, and there's no way to stop it. The analogue in this case would be if your grandmother came to visit, and watched "Lawrence Welk" reruns on Public Television. Then, ContentWise kept recommending shows with accordion players. Both Neptuny and Orca assured us that viewers can go in and delete shows from their watching history. Nevertheless, I'll believe it when I see it.

As with search, no one has found the "holy grail" of 21st century video discovery yet, and the company (or companies) that do will have a tremendous opportunity to license their technology.

No comments: