Recommendations are not really a framework thing, though there are open source libraries (like collaborative filtering libraries) you could use.
Recommendations are not trivial - if done badly they can really suck. And there's a whole bunch of computer science and mathematics behind different recommender systems. I don't know much about this (even I'm in the learning process), but there's no recommendation system you can "plug and play". Firstly, what are you trying to recommend? Secondly, you need to define the relationships between items that are being recommended and users. For example, if user likes items A, B and C, and a lot of users like both A, B and D, you probably want to recommend D (this is collaborative filtering, like Amazon or Last.fm). In this case you build huge graphs of what items are liked by what users, and based on common nodes arrive at some score of "influence" of one user on another, and use it to drive recommendations. This is a very high level view - read up collaborative filtering, find some open source libraries to play around. Alternatively, a simpler implementation would need you to arrive at a score for various items - and the score would be a sum of various criteria as per your usage. Suppose usage = viewing an item + buying an item + reviewing an item, then you give a weightage to each of these "actions", come up with a score for each and every item for each and every user, and sort them in reverse order of score for recommendation. Basically this implementation doesn't factor the popularity of an item in your system, but is driven by a user's profile extracted by tracking his usage. Here again you need to build it yourself, and keep revising the weightages of the actions till users start liking your recommendations. Hope some of this made sense! On Sep 12, 4:48 am, Francisco Costa <[email protected]> wrote: > Hi, > I'm building this social web app and now I'm considering implement > some recommendations to the users based on their usage or profile. > Can someone point some available systems that go well with web2py? > Would you consider develop your own recommender system or use an > existing one?
