cf-based-recommendation's Introduction

CF-Based-Recommendation

Recommendation System based on Collaborative Filtering

README

To compile the program you need java. To create class file, execute the command

javac recommender.java

Once the class file is formed, different configuration of recommender system can be run. If you want to run cross-validation you will have to mention second argument as "Best" as undermentioned, else you can specify any one type of similarity type as second argument. Also since java is row-major, both Item-based and Combined-based program became slow to ensure generality. User-based predictions with predefined similarity could take upto 45-mins on quad-core i7 system, whereas Item-based and Combined-based could take upto 2 hours. If you run cross-validation, then the code will use the entire ratings.csv file, for the entire 8,00,000 entiries and Item-based system could take upto 12 hours to run.

2.1) User-Based I) To run user based recommender system with Pearson similarity execute -

java recommender User Pearson

II) To run user based recommender system with Cosine similarity execute -

java recommender User Cosine

III) To run user based recommender system with Jaccard similarity execute -

java recommender User Jaccard

IV) To run user based recommender system with Best similarity, using K-fold cross-validation execute -

java recommender User Best

2.2) Item-Based I) To run item based recommender system with Pearson similarity execute -

java recommender Item Pearson

II) To run item based recommender system with Cosine similarity execute -

java recommender Item Cosine

III) To run item based recommender system with Jaccard similarity execute -

java recommender Item Jaccard

IV) To run item based recommender system with Best similarity, using K-fold cross-validation execute -

java recommender Item Best

2.3) Combined User-Item Based I) To run the combined user-item based recommender system with Jaccard similarity execute -

java recommender Combined None

Recommend Projects