2010/3/31 Christoph Hermann <[email protected]>:

>> You could modify the code to use this info for your purposes if you
>>  like.
>
> I'll try to implement this.
> What i would do is:

Look at lines 116-123 of GenericRecommenderIRStatsEvaluator. Just
stick in logic to fill in 'relevantItemIDs' in whatever way you want.
Here, you'd pick IDs of items whose preference was set after some
date. The rest should still work out.


> Well if you take the items as web pages it makes more sense, doesn't it?

Hard to say. I imagine your set of user-to-webpage associations is
relatively dense, that you know about most meaningful connections of
this type. If that's true, then I think the test is more meaningful.
For more sparse data, I think it falls down more.


> I could think of a few cases where evaluating a boolean recommender with
> two time separated dataModels makes sense.

To generalize this... I guess we could add some kind of
"RelevantItemChooser" interface to let a caller stick in some notion
of relevance. It could be implemented to choose based on date or
whatever. Would that work for your case.

Reply via email to