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

[Ryan Curtin]

On Tue, Mar 08, 2016 at 04:53:59PM +0530, Nilabhra Roy Chowdhury wrote:
> Hello,
> 
> I am a senior year student in computer science and engineering from India.
> I have been working on machine learning problems for past two years and
> have come across methods such as SVD and Weighted Nearest neighbors for
> collaborative filtering techniques. Once I have also used regression to
> interpolate the values for collaborative filtering.
> 
> I am willing to contribute to this project as I want to implement the
> algorithms myself and also explore more advanced techniques such as Field
> aware Factorization Machines (FFM). I will be building mlpack in my system
> shortly. Please let me know if there is any preference of literature and if
> I am required to do some bug fixing before being able to contribute.

Hi Nilabhra,

It sounds like you are already familiar with matrix factorization
approaches to collaborative filtering, so if you have a good
understanding of the existing literature (like Koren's paper and other
related ones [1]), then it probably makes more sense to focus on the
existing codebase and what it can do.

So I would suggest spending some time looking at the code in
src/mlpack/methods/cf/, src/mlpack/methods/amf/,
src/mlpack/methods/regularized_svd/, and possibly
src/mlpack/methods/quic_svd/ to learn about the existing CF system, so
that you can figure out how the API for your proposed techniques would
look.

We already have some SVD techniques implemented (see the AMF code), but
weighted nearest neighbors and FFM might be interesting.  There's lots
of fast code for nearest neighbor search in mlpack already, so that may
be useful to look into also.

Let me know if I can clarify anything.

Thanks,

Ryan

[1] https://datajobs.com/data-science-repo/Recommender-Systems-[Netflix].pdf

-- 
Ryan Curtin    | "If you understood everything I said, you'd be me."
ryan at ratml.org |   - Miles Davis