I was browsing through some online book stores in India and I was not really happy with the recommendations they made. The recommendations were naive, obvious and not interesting. Reason?
When I use Google, I think in terms of "context" not "Object". For example I Type 'books + different algorithms on web intelligence" this is a context that I am thinking and I might not have a certain object in mind as in a book title or a publication journal. I just have a context I am interested in and I do the search, get the result and proceed with what ever. Now coming back to these search portals of e-commerce sites, local listings, classifieds etc. They are a repository of 'objects'. And they enable search on these objects while I have a context. Now this does not really fit well, I mean I look for a context in such an repository. And then they making recommendations depending on the kind of search I frequently make, or the items I click. here whats missing is a set of few other informations.
Fact 1: I would not have searched just their portal or listing, I certainly would have searched else where (similar portals) parallely.
Fact 2: Probably I would have searched in google too and it would have returned me results for my context.
Fact 3: What else I might have in mind when I types those key words?
Lately what i have observed is Google has spoiled the way we search by providing the luxury of intelligence in it. People tend to type in direct english (or other language) sentences as if they are conversing and expect exact result.
Then how to leverage on that piece of information?
Why can't our recommendation engine browse through the recent history of my browser, collect my search results and habit on the similar context and objects? We can actually use that data and the listing by the other portals, google before making a recommendation and precisely zero in on what I need.
That way the e-commerce portal can map more correctly from the context I am searching to the object they have and are selling. And I will come back to them as I find more suitable and exact results with them.
Just a random idea.