mahout-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sean Owen <sro...@gmail.com>
Subject Re: integrating databases
Date Tue, 01 May 2012 08:38:47 GMT
(I think the question is more blending two data sources than two
recommenders.)

On Tue, May 1, 2012 at 9:37 AM, Manuel Blechschmidt <
Manuel.Blechschmidt@gmx.de> wrote:

> Hi Amrhal,
> combining data for a recommender from two data sources is current
> research. Search for ensemble learning or blending learners. Basically the
> combination of multiple learners is called blending. (
> http://pragmatictheory.blogspot.de/2008/07/blending-101.html). Blending
> different models is currently one of the most important features to get
> these great accuracy's like in the NetFlix Price or blending offline and
> online learning methods to do something in semi realtime (Fast Online
> Learning through Offline Initialization for Time-sensitive Recommendation -
> http://users.cis.fiu.edu/~lzhen001/activities/KDD_USB_key_2010/docs/p703.pdf
> ).
>
> I recommend the chapter Combining Multiple Learners from Introduction to
> Machine Learning. (
> http://www.amazon.com/Introduction-Machine-Learning-Adaptive-Computation/dp/026201243X/
> )
>
> As far as I know there are currently no ensemble or blending learners in
> Mahout. Here you find a block diagram how such a data model could look like:
>
> https://source.apaxo.de/svn/semrecsys/trunk/doc/images/SemanticRecommenderDataModel.pdf
>
> In the block diagram you see 3 data source: explicit ratings, implicit
> views and explicit purchases. This data model should be optimized to sell
> as much products as possible. So a customer can interact in up to 3 ways
> with a product view it, rate it, and buy it. You could call these different
> contexts and then use different contexts for learning (see Workshop on
> Context-Aware Recommender Systems 2009/2010/2011). e.g. you could use
> Tensor Factorization etc.
>
> Further you have to differentiate between the data model and the learner
> (recommender). I just describe how to make the learner more accurate and
> use data from different sources. If you just keep the same data for the
> same context in your Oracle database then it won't be very complicated to
> integrate it with the MySQL database.
>
> Have a great week
>    Manuel
>
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message