[mlpack] Alternatives to neighborhood-based collaborative filtering for GSOC 2016

[Ryan Curtin]

On Fri, Mar 11, 2016 at 01:22:48AM +0530, b venk wrote:
> Respected Sir ,
> I am Venkatesh , a Computer Science student from India. I am interested in
> contributing to the project "Alternatives to neighborhood-based
> collaborative filtering".
> I have been reading about a few matrix factorization models for the same.
> One of the models that I  propose to implement is the  modified SVD
> ,(SVD++) that makes use of implicit feedback using classes from mlpack::svd
> namespace for recommendation.
> The method is well described here (here) .
> <http://delivery.acm.org/10.1145/1410000/1401944/p426-koren.pdf?ip=182.75.45.1&id=1401944&acc=ACTIVE%20SERVICE&key=045416EF4DDA69D9.CC8C54AC3A0AC65D.4D4702B0C3E38B35.4D4702B0C3E38B35&CFID=760521007&CFTOKEN=40057613&__acm__=1457613157_e2e7b93b4e1710e3a5289515271b1136>
> 
> Also I propose to implement models that use time factors , (like the rating
> drop of a movie  over time etc). Other models that would be implemented
> includes the weight based  kNN (that is mentioned in the project
> description). The data that I would be using to test my models would be
> from Netflix
> and Kaggle. I have been working with C++ for the past 5 years and am good
> with  implementation. I request for your valuable suggestions / additional
> methods.
> 
> I also thank you for contributing to this wonderful project - mlpack.

Hello Venkatesh,

Thanks for getting in touch.  SVD++ and the other techniques you've
suggested would be interesting.

If you are thinking about adding time support, we need to think about
how the user will give their data.  Right now, the user passes an
arma::mat (or arma::sp_mat) of recommendations to the CF class.  But if
there is time information, what will they pass instead?  This is
something that needs thought and discussion to make sure that we can
come to the best decision that matches best with the rest of the mlpack
code.

I hope this is helpful; let me know if I can clarify anything.

Thanks,

Ryan

-- 
Ryan Curtin    | "What? Facts?"
ryan at ratml.org |   - Joe Cairo