eTech05: Ask Jeeves Alpha

One of the meta tracks here at eTech seems to be “a peak inside modern software research labs”. Today, we heard from Ask Jeeves.

Ask Jeeves provides an alternate search engine. In their words, not better than google, just different.

Ask Jeeves Alpha, a double-secret codename for their research group. They will eventually have a public-facing site.

New technologies:

* Related Topics
* Clustering of Related Topics
* New image search

Teoma, their search technology, tries to a more category/conceptual based lookup. Sounds a bit like the beginnings of semantic search?

Uses a hubs and authorities approach to perform the breakdown and analysis. This is layered with traditional results to provide the final answer.

He’s talking about the benefits of their ExpertRank system, and mentions “expert validation”. I’ll be listening to learn how this works, and if the experts are human, how it will scale.

In his example, Teoma returned sites with content, and google returned sites with more links to content. Interesting.

Discusses the iterative nature of most searches and how they can use the Related Topics feature to provide better results. Plus, google has a related topics search as well, so … ahh, he addresses that. Google still doesn’t return conceptual or topic results, just keyword results. I guess the key here is, if it works, Teoma/Ask Jeeves intends to provide topic/concept search. That would be a good thing!

(yeah, this talk is a mix of technical and marketing info. Still, the point is that search is not “done” yet; there’s plenty of room for improvement. Also, what’s w/Britney? She was used as an example in the Long Tail talk earlier today.)

Turning to their new image search, it seems they are not only spidering based on metadata, but are also using communities (unspecified how) to derive context for images. Hmmm …

The founder of Teoma provided a brief background on their technology and a demonstration of these technologies. Teoma was founded on the concept of topic/conceptual based search, and expanded into the use of “communities” for deriving context.

The clustered search results shown in the demo did look very powerful; I look forward to the day when this is released from their labs. It will be interesting as 3rd parties compare these new technologies with ones from competitors such as MSN, Yahoo and google.

The talk ended with a brief preview of their new image search. More info is available on their blog, particulary in this post.