Movie Recommendor System

Dataset Overview

Original Dataset (Netflix Challenge dataset): The dataset is divided into three sets.

  1. Training set
  2. Qualifying set
  3. Probe set

The difference between the qualifying set and the probe set is that the ratings for the qualifying set are withheld by Netflix and they are not released to public. The probe set, however, is created to help the participants test the accuracy of their methods before submitting the results for the qualifying set to Netflix.

Below is the breakdown of the data in numbers.

• Total movies across the entire dataset: 17770 Movie IDs range from 1 – 17770 with no gaps

• Total Customers/Users included in the entire dataset: 480189 Customer IDs range from 1 – 2649429, with gaps

• The rantings are on a scale from 1 – 5

• The date of rating is also provided in the dataset. It is in the format YYYY-MM-DD

Results

Method 1 -- RMSE: 2.213309 and MAS: 1.846437

Method 2 (with part of the dataset) -- RMSE: 1.08041 and MAS: 0.80641

Method 2 (with the entire dataset) -- RMSE: 1.060477 and MAS: 0.77795

Method 4 (with part of the dataset) -- RMSE: 0.368629 and MAS: 0.120844

For details of all the four methods, see the full report and slides avaialble in this repo.