CIKM'14 - A recommendation method based on multi-armed bandit framework with clustering
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
.settings
Data
Input
Output
bin/com/smu
libs
src
.DS_Store
.Rhistory
.classpath
.project
README.txt
cikm.jar

README.txt

DYNUCB: Dynamic Upper Confidence Bound
--------------------------------------

INTRODUCTION: 

- This is an implementation in java for the paper "Dynamic Clustering of Contextual Multi-Armed Bandits" (Nguyen & Lauw, CIKM 2014)


The folder includes:
- Data: for importing the Delicious and LastFM database (using MYSQL(usr: 'root', pw: '') for connection).
- Input: the norm matrix after reducing dimensions.
- Output: the results is separated into N-tries (N threads) for taking the average.


*** Run from terminal:
java -Xmx4G -jar cikm.jar [args]

with arguments in order below:
- 1st: algorithm selection (1-LinUCBSIN; 2-LinUCBIND; 3-DynUCB)
- 2nd: dataset selection (1-delicious; 2-lastfm)
- 3rd: the number of running times/threads of the selected algorithm (e.g. 10)
- 4th: the number of iterations (e.g. 50000)
- 5th(optional): the number of clusters (default 16-clusters) in case of choosing DynUCB algorithm


Examples:
- Running DynUCB algorithm on Delicious dataset with 16 clusters, 10 threads, 50000 iterations.
java -Xmx4G -jar cikm.jar 3 1 10 50000 16


HOW TO CITE:

If you use DynUCB in your research, please cite the paper with the bibtex format below:

@inproceedings{nguyen2014dynamic,
  title={Dynamic clustering of contextual multi-armed bandits},
  author={Nguyen, Trong T and Lauw, Hady W},
  booktitle={Proceedings of the 23rd ACM International Conference on Conference on Information and Knowledge Management},
  pages={1959--1962},
  year={2014},
  organization={ACM}
}