The reboot of that old mr engines in Mahout-Samsara is what we call Correlated
Cross-Occurrence (CCO) this is the core of a mutli-modal recommender engine
that can use almost any information about the user, context, or items to make
recommendations. It is the first Open Source version of this algorithm as far
as I know.
This http://mahout.apache.org/users/algorithms/intro-cooccurrence-spark.html
describes the core engine in Mahout.
An end-to-end implementation with event ingest, training, and scalable
deployment under Apache 2 license is here:
https://github.com/actionml/template-scala-parallel-universal-recommendation/tree/v0.3.0
Notice many improvements over the old (soon to be deprecated) mr implementation:
realtime queries—fast responses
uses realtime usage data to capture most recent user behavior—allows person
recs to anonymous new users
uses as much of the user’s clickstream and context as makes sense—the
multi-modal cross-occurrence part
can use preferences for categories, tags, genres. brands, location. devices,
etc—more multi-modal cross-occurrence
We have recently begun testing the multi-modality features against public data
and will be publishing some very encouraging results. One interesting finding
is where we took user “likes” and “dislikes” from reviews on rottentomatoes and
found that “dislikes” along with genre preferences gave us a 23% increase in
mean average precision over using “likes” alone. Yes, that means dislikes may
predict likes. Unless a recommender is multi-modal it can really only use one
user action—“likes” in this example. So ALS is not multi-modal.
> On Feb 19, 2016, at 1:30 AM, Lee S wrote:
>
> @Adi this link is for als algorithm, not the item-based implementation.
>
>
> On Fri, Feb 19, 2016 at 1:09 PM, Adi Haviv wrote:
>
>> collaborative filtering -
>> https://codeascraft.com/2014/11/17/personalized-recommendations-at-etsy/
>>
>> On Fri, Feb 19, 2016 at 8:46 AM, Lee S wrote:
>>
>>> Hi:
>>> Does anybody know which paper the mr algorithm is based on?
>>>
>>
>>
>>
>> --
>> Adi Haviv.
>>