SVDFeature

From SVDFeature
(Redirected from Main Page)
Jump to: navigation, search
SVDFeature
A Toolkit for Feature-based Collaborative Filtering and Ranking

Project Description

Collaborative filtering is among the most promising approaches to recommender system. In real world scenarios, many types of auxiliary information are available besides the user preference matrix and the use of them may enhance the prediction accuracy. For example, information such as time and location can help better predict users' preferences based on the contexts, while information such as users' ages and gender can help make better prediction on the basis of user profiles. From the viewpoint of learning, the use of auxiliary information can overcome the sparsity of the matrix data.

SVDFeature is a toolkit designed to efficiently solve large-scale collaborative filtering problems with auxiliary information. Unlike traditional engineering approaches for collaborative filtering which requires writing specific solver for each algorithm, SVDFeature solves a general form of CF problems thus allow develop new models just by defining new features. The feature-based setting allows us to include many kinds of information into the model. Using the toolkit, we can easily incorporate information such as temporal dynamics, neighborhood relationship, and hierarchical information into the model.

Features

  • Handling Big data in a single machine: The toolkit buffers the training data on disk thus memory cost is invariant to training data size, the code is optimized for good performance.
  • Building state-of-art CF via feature engineering: Many variants of matrix factorization can be described in feature-based matrix factorization. One can try new approaches by generating corresponding features, and no modification of code is required.
  • Fast Training Speed: The toolkit implement efficient training algorithm for user feedback information, the vector operation is optimized by sse. We use a multi-thread pipeline to do disk pre-fetching.
  • Supported objective functions: regression, classification, learning to rank

News

  • 2012-12: Our paper about SVDFeature has been published by Journal of Machine Learning Research, machine learning open source software track

Achievements

  • 2012-06-02: We achieved the 1st place in track1 of KDD Cup 2012 Leaderboard, Online Publication Scripts Beijing, China, 2012.
  • 2011-07-01: We achieved the 3rd place in track1 of KDD Cup 2011 Leaderboard. Inner Peace is the SJTU-HKUST joint team. We report the best single method in track1.

Getting Started

Google Group

Open Thoughts

Extensions

SVDFeature can be extended to support other models that utilize auxiliary information, we reuse the input module of SVDFeature. The underlying model does not have to be feature-based matrix factorization.

Downloads

  • Older Versions

Publication

  • Tianqi Chen, Weinan Zhang, Qiuxia Lu, Kailong Chen,Zhao Zheng, Yong Yu. SVDFeature: A Toolkit for Feature-based Collaborative Filtering. Journal of Machine Learning Research. 13:3619-3622, 2012. (PDF) (BibTex)

People

Links

Personal tools
Namespaces

Variants
Actions
Navigation
Toolbox