In this repo, I implemented AutoRec, a collaborative filtering model proposed in the paper Autoencoders meet collaborative filtering with Keras and tuned hyperparameters of this model using the validation set.

AutoRec is an autoencoder framework for collaborative filtering(CF). There are two variants of AutoRec depending on two types of inputs: item-based AutoRec (I-AutoRec) and user-based AutoRec (U-AutoRec). I implemented I-AutoRec in this repo. I-AutoRec, the input of the model is the item-interaction vector. For item i, the item-interaction vector of i is the ith column of the rating matrix.

I evaluated AutoRec on 100K Movielens dataset.

I tune the following hyperparameters using validation data: number of units in hidden layer=100, the activation function of hidden layer=tanh, and the optimization method = adam.

Result of tuned autorec on Test set	Results for different number of hidden units

Results for different activation functions	Results for different optimization methods

git clone https://github.com/mshadloo/Recommendation-system.git
cd Recommendation-system
chmod +x data.sh && ./data.sh
chmod +x run.sh && ./run.sh