I am using mahout core to implement several recommender functions. Among others, I want to implement a "trending products" recommendation function. Usually I would use something like Z-score, but I want to stick to things mahout provides and I could not find anything in the libabries.
Does anybody know, how to implement a trending function using mahout?
EDIT: What I try to accomplish: I want to implement a general purpose recommender engine, that should offer a) recommendations based on the the user history and b) recommendations about currently trending products. therefore products that got more clicks in the last N days then usual.
This question is only targeting b)
Not sure exactly what you're trying to do, but you might consider looking at http://commons.apache.org/proper/commons-math/javadocs/api-3.5/org/apache/commons/math3/distribution/NormalDistribution.html
The distributions in
org.apache.commons.mathare pretty good and that's why we didn't recreate them.UPDATE OK, you're getting off into some math theory and should probably ask this on stats.stackexchange.com or something like it as it is relative to statistics, not the programming implementation. What you're talking about is a sort of Delta recommender. You could bootleg it by using the Mahout CCO reccomender and having a seperate "feature" (or matrix) of items and "clicks in the last 7 days". There are multiple ways to skin this cat though, for more info on how to implement in Mahout please join the dev or user mailing lists, as there will be some active community discussion on the best method to pursue.
https://mahout.apache.org/general/mailing-lists,-irc-and-archives.html