This is a reommender system building with Movielens movie ratings.The whole system segments users as New User and Exsiting User and apply different analysis on each segment inspired by Scikit-Learn

For exsiting user, we used collaborative filtering analysis to get the users' protential preferences.

The preprocessing process is the process performing matrix factorization.In this project, I used two methods to apply Sigular Value Decomposition:

• Stochastic Gradient Descent
• Alternating Least Squares

The measures to do the evaluation is :

• Root Mean Square Error
• Mean Absolute Error However, in the methods comparison, I mainly used RMSE

The prediction process is to predict the rating after performing matrix facotrization and the results can be used to recommend movies. Here I used KNN algorithm to do so.

Benchmark

Training time:

 SVD : 72.5138s
 ALS : 21.6902s

Evaluating RMSE:

          Training Error         Testing Error
SVD         0.002393               0.018424 
ALS         0.655201               1.308651
SVD-KNN     0.000251               0.003521
ALS-KNN     0.001760               0.007042

Analysis