Section: New Software and Platforms

Collaborative Filtering Network

Keywords: Recommender system - Neural networks - Deep learning

Functional Description

Recommendation systems advise users on which items (movies, musics, books etc.) they are more likely to be interested in. A good recommendation system may dramatically increase the amount of sales of a firm or retain customers. For instance, 80% of movies watched on Netflix come from the recommender system of the company. Colaborative Filtering (CF) aims at recommending an item to a user by predicting how a user would rate this item. To do so, the feedback of one user on some items is combined with the feedback of all other users on all items to predict a new rating. For instance, if someone rated a few books, CF objective is to estimate the ratings he would have given to thousands of other books by using the ratings of all the other readers.

The following module tackles Collaborative Filtering tasks by using a novel approach based on neural networks (sparse denoising autoencoders). In a few words, the module lets the user train neural networks to predict unknown entries in a history files.

The input files are classic csv files. The output files can either be the full matrix of ratings and/or the network weights. The root mean square error is computed to assess the quality of the training.

This module is based on Lua/Torch Framework. It works on both CPU/GPU and it is multithreaded.

• Contact: Florian Strub

• URL: https://github.com/fstrub95/Autoencoders_cf