Wednesday, May 26, 2004

Infospace on personalized search

Search Engine Lowdown has an interview with Arnaud Fischer from Infospace (owners of Dogpile and WebCrawler).

An interesting little snippet on personalized search:
    Monitoring navigation behavior at a user-level could conceivably be the basis to developing an understanding of users' individual interests over time, in essence personalizing the equivalent of Google's PageRank scores. If you consistently browse music-related content, search engines should become smart enough to understand that your query "Prince" most probably relates to the singer than to the royal family. Personalizing search relevancy algorithms presents some major scalability and performance challenges, though. It takes days, if not weeks to process link analyses and compute authority scores for individual Web sites after a crawl.
Update: Another interesting quote from this article that I missed the first time I looked at it.
    Analyzing click popularity at an aggregate level along IP-associated parameters could be leveraged to extrapolate personalized ranking for clusters of users exhibiting similar behaviors. This technique would not be unlike Amazon's implementation of collaborative filtering technology.
Arnaud Fischer is definitely on the right track here. It's quite tricky to get this kind of thing right though. It'll be interesting to see if Infospace makes an attempt.

No comments: