likelike

An implementation of locality sensitive hashing

Overview

Likelike is an implementation of LSH (locality sensitive hashing) on Hadoop. This program can be used for the nearest neighbor extraction or item recommendation in E-commerce sites. Currently Likelike supports only Min-Wise independent permutations. Min-Wise independent permutations is applied to the recommendation of Google News (Das et al. 2007).

Usage

Begin with the Likelike quick start page (QuickStart) which provides the information on the installation and tutorial with small input files. For detailed usage, please visit the Usage page.

News

• 2015-02-03 support Hadoop v2.4
• 2013-01-01 port the code into Github
• 2011-09-26 gave a presentation in the Hadoop conference Japan Fall (slides)
• 2011-08-22 likelike 0.3.0 released switched the build system from Ant into Maven.
• 2010-05-06 likelike 0.2.0 released

Requirements

• Java 1.７.0 or greater
• Hadoop 2.4 or greater

Applications

Likelike can be applied to following tasks.

• item recommendation in E-commerce sites
• similar documents (or images) extraction
• news recommendation from click through log data

To do

• provide detailed usage page
• provide server function
• support feature selection
• performance tuning

Author

Takahiko Ito <takahiko03 --at-- gmail.com>

References

• Edith Cohen. Size-Estimation Framework with Applications to Transitive Closure and Reachability. Jour. of Computerand System Sciences. 1997.
• Abhinandan S. Das, Mayur Datar, Ashutosh Garg and Shyam Rajaram, Google news personalization: scalable online collaborative filtering, In: Proc. of WWW conference, 2007.
• Indyk Piotr and Motwani Rajeev. Approximate Nearest Neighbors: Towards Removing the Curse of Dimensionality. In: Proc. of Symposium on Theory of Computing. 1998.