Prastava - An open source, ruby-based generic recommendation system

Prastava, the open source Recommendation system

This is the home page of the Prastava project, which is a generic, ruby-based open source recomendation system. This project can be used to generate recommendations for any item user database.

You can generate recommendation based on three recommendation schemes: collaborative filtering, content-based filtering and a hybrid of both.

If you're wondering why we named our project prastava, the word prastava means "recommendation" in Hindi.


Download

Download Prastava version1.0( Library needed - Ferret gem )

Documentation

Prastava documentation

Web-based demonstration

In this web based demonstration, you can choose to generate recommendations on either the famous MovieLens database, or provide your own data set to generate recommendations on.

Database to use:

   Movie Lens Database

Change the '0' to your rating. Edits to the titles or rows will have no effect. The text below defaults to a user that likes "Kolya" with rating '5', "L.A. Confidential" with rating '4' and "Legends of the Fall" with rating '3', and all other movies unseen. Edit to get different results in our demo. Ratings should be between 1 (awful) to 5 (must see). You should rate at least 15 movies to generate better recommendations.



     Your Own Toy Data

Enter your own data in the following triple format: "Item ID/Name" "User ID/Name" "Rating (in the range 1-5)". Each of these fields should be seperated by a space and enclosed in quotes. Give each new rating on a seperate line. For example, "Slumdog Millionaire" "Tarun" "5" means that the user "Tarun" gave a rating of "5" to item "Slumdog Millionaire"


"User ID/Name" (in quotes) for which you want to system to give recommendations.

Number of Recommendations to generate:                                     




Filtering Technique:

   Collaborative Filtering
          Number of Nearest Neighours   
   Content Based Filtering

   Hybrid Filtering




Collaborative Filtering Algorithm:

   Cosine
   Pearson






Note: On the data that you provide we only perform collaborative filtering as the items' content isn't currently given.

Group Members


The Prastava Team of WING
Created on: Mon 13:07:15 2009 | Version: 1.0