svd++

SVD++ refers to a matrix factorization model which makes use of implicit feedback information. In general, implicit feedback can refer to any kinds of users' history information that can help indicate users' preference.

MODEL FORMALIZATION

The SVD++ model is formally described as following equation:

where  is the set of implicit information( the set of items user u rated ).

GENERAL FORMALIZATION FOR USER FEEDBACK INFORMATION

A more general form of utilizing implicit/explicit information as user factor can be described in following equation

Here  is the set of user feedback information( e.g: the web pages the user clicked, the music on users' favorite list, the movies user watched, any kinds of information that can be used to describe the user).  is a feature weightassociates with the user feedback information. With the most two common choices: (1)  for implicit feedback, (2)  for explicit feedback.

（一直搞不清楚上面公式当中yi到底是什么，现在清楚了，yi就是上面所写出来的隐式反馈！）

LEARNING

SVD++ can be trained using ALS.

It is slow to train a SVD++-style model using stochastic gradient descent due to the size of user feedback information, however, an efficient SGD training algorithm can be used. [1] describes efficient training with user feedback information in section 4

LITERATURE

• Yehuda Koren: Factorization meets the neighborhood: a multifaceted collaborative filtering model, KDD 2008, http://portal.acm.org/citation.cfm?id=1401890.1401944

IMPLEMENTATIONS

• The GraphLab Collaborative Filtering Library has implemented SVD++ for multicore: http://graphlab.org/pmf.html
• SVDFeature is a toolkit designed for feature-based matrix factorization, can be used to implement SVD++ and its extensions.
• LibFM can also be used to implement SVD++
• wooflix is a (not very fast) Python implementation of SVD++
• MyMediaLite: SVD++ source code on GitHub; see also [2]