Monday, October 03, 2005

Search as a dialog and personalized search

John Battelle posted an interesting quote from Gary Flake (formerly of Yahoo Research, now at Microsoft) on how people search:
[Gary] wished searchers were more ... sure of what they wanted, and willing to engage in a dialog with the search engine. Most, it turns out, are not.
If there is going to be a dialog, it needs to be something easy, something that helps searchers without requiring any effort on their part.

Current search engines treat each search is independent, ignoring the valuable information about what you just did, what you just found or failed to find. Paying attention to that history should allow search to become more relevant, more useful, and more helpful, all with no effort from searchers.

In his new book, John has another quote from Gary Flake about using the wisdom of the crowds to improve search results:
"You can learn a lot by watching the statistical patterns of search usage and leveraging that in algorithms," notes Gary Flake ... "We use a very large corpora (body of data) to identify sets of tactical and grammatical properties of language."

The result: search has the potential to get better and better, the more people use it.
While Gary is talking about the kind of techniques used for automated spelling correction, machine translation, and question answering, the techniques can also be used to learn what individual people think is and is not relevant, to pay attention to the history of what people have done, to do personalized search.

1 comment:

Scott said...

My ardent hope is that Google's personalized search capabilities will lead to some nirvana-like search experience. I've benefited from the Personalized Search features in quickly finding content I recently searched... I hope evetually Google will use this data to help with recommendations a la Findory.

Also, as you noted, it should be able to look at my recent search patterns to help do some pre-filtering.

(What I wouldn't give to be able to run some queries against Google's databases... would be interesting to see aggregated stats on a wide variety of search patterns.)