Question

By dynamic data rating I mean a time-based recommendation system.

One example use case for movie recommendation, the recommendation engine read user history movie watching data and find out that user likes watching action movie on weekend, the engine should rate with higher score for action movie.

However the same type of movies might rate with lower score during weekday since historical data might suggest that user likes horror movie during weekday.

In other words, the same piece of historical data score differently, depending on when the recommendation happens.

Can we achieve this with Mahout?

Thank you.
George

Answer 1

Yes and no. No in the sense that there is no algorithm implemented in Mahout that uses time directly. Yes in the sense that there are probably enough hooks that you can add such logic without rewriting the implementation entirely.

The most directly relevant hook (for non-distributed recommenders) is IDRescorer. This would let you boost or demote items based on whatever external logic you like. It could be time-based.