Java-Based Context-aware Recommendation Library
Switch branches/tags
Nothing to show
Clone or download
Latest commit c836a09 Aug 18, 2018



CARSKit ( is an open-source Java-based context-aware recommendation engine, where it can be used, modified and distributed under the terms of the GNU General Public License. (Java version 1.7 or higher required). It is specifically designed for context-aware recommendations.

GPL License

CARSKit is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License (GPL) as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. CARSKit is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with CARSKit. If not, see

Architecture and Design

CARSKit Structure


  • Traditional recommendation algorithms: The implementations of those algorithms (such as UserKNN, BiasedMF, SVD++, SLIM, etc) are from LibRec-V1.3. Those algorithms can be used in two ways: 1). run a traditional recommendation algorithm directly on the context-aware data set to compete with the context-aware recommendation algorithms; 2). run contextual recommendation algorithms based on transformation, e.g., run a traditional recommendation algorithm after data transformation (e.g., by item splitting).
  • Context-aware recommendation algorithms: CARSKit simply divides it into two categories: Transformation Algorithms and Adaptation Algorithms. The transformation algorithms try to convert the mulidimensional recommendation problem into traditional 2-dimensional problem, so that the traditional recommendation algorithms can still be used. In contrast to those algorithms based on transformation algorithms, the adaptation algorithms focus on their effect on building algorithms by adapting to the multidimensional rating space. Most of the algorithms inside belong to the contextual modeling algorithms, such as Context-aware Matrix Facatorization (CAMF), Tensor Factorization (TF), Contextual Sparse Linear Method (CSLIM), Factorization Machines (FM), etc.


Please cite the following papers if you use CARSKit in your research:

  1. Yong Zheng, Bamshad Mobasher, Robin Burke. "CARSKit: A Java-Based Context-aware Recommendation Engine", Proceedings of the 15th IEEE International Conference on Data Mining (ICDM) Workshops, pp. 1668-1671, Atlantic City, NJ, USA, Nov 2015

If you use CARSKit in your research papers, please let us know. And I will create a list of publications using this toolkit.

Downloads (Latest)

Data Sets

A list of context-aware data sets can be found here:

User's Guide

The guide can be found and downloaded from
Note: if there are missing values or EMPTY values in the context information, you should mark those values as NA (i.e., Not Available) which will be viewed as a special context condition in the data.


I would like to show our gratitude to Dr. Guibing Guo (the author of LibRec) for his comments and suggestions on the development of CARSKit.

Who uses CARSKit

  • Universidad del Bío-Bío, Chile
  • Pontificia Universidad Católica de Chile, Chile
  • Shanghai Jiao Tong University, China
  • South China Agricultural University, China
  • University College Cork, Ireland
  • Polytechnic University of Bari, Italy
  • Gdansk University of Technology, Poland
  • Universidad Autónoma de Madrid, Spain
  • University of Manouba, Tunisia
  • University Tunis El-Manar, Tunisia
  • National Institute of Applied Science and Technology of Tunis, Tunisia
  • DePaul University, USA
  • University of California - Santa Barbara, USA

[Contact me if you want to be listed here]

Release Notes

2016/05/04, Version 0.3.0

  • BugFix in some algorithms
  • Add more contextual filtering algorithms: ExactFiltering, DCR and DCW
  • Add "-fullstat" in configuration which allows the kit to present more details on data statistics
  • Add "-minlenu" and "-minleni" to setup "recommender=usersplitting -traditional biasedmf -minlenu 2 -minleni 2"
  • Update corresponding JavaDoc and JAR library

2016/04/18, Version 0.2.4

  • BugFix in Tensor Factorization
  • Add a contextual PreFiltering algorithm: SPF (Semantic PreFiltering)
  • Update corresponding JavaDoc and JAR library
  • Change the evaluation process when you set a rating threshold for TopN recommendation

2015/11/09, Version 0.2.0

  • add Tensor Factorization (CPTF) into CARSKit
  • update API documents and the configuration file setting.conf
  • release user's guide, "A User's Guide to CARSKit"
  • optimize the coding

2015/09/25, Version 0.1.0 (Initial Release)