I just finished watching a Great Tech-Talk on how Google news recommends relevant news items one might be interested in and
I thought it would be nice to share this video on my blog.
Please note the talk was given in 2008, but is quite informative None the less.
Some Takeaways:
They mainly use collaborative algorithms algorithms family for coming up with stories to recommend.
Using this not only can they leverage google’s large user base that they have to their advantage, but it also makes algorithms domain independent, so the same algorithms can also be used to some other domain like google video, etc.
They use weights on various applicable methods to come up with a net weighted score for all the stories.
Some of the factors influencing news recommendation:
- Stories which users who have similar interest as yours have clicked [MinHash,PLSI]
- Stories that you have co-visited with the stories you clicked.
Recommendation are category specific, so your news clicks in entertainment section won’t affect items recommended in technology section.
Author also talks about how to make these algorithms scalable enough to run them effectively, which is mostly achieved by using Map-Reduce , a well known distributed computing standard.
To people genuinely interested on this topic, I would suggest reading another great book on the subject – programming Collective Intelligence: Building Smart Web 2.0 Applications
A great read for the researchers and the newbies alike.
Ashish Yadav's Weblog 