Recommender Systems

This directory has a set of examples of how to build various kinds of recommender systems using MXNet. The sparsity of user / item data is handled through the embedding layers that accept indices as input rather than one-hot encoded vectors.

Examples

The examples are driven by notebook files.

• Matrix Factorization: linear and non-linear models
• Deep Structured Semantic Model (DSSM) for content-based recommendations

Negative Sampling

• A previous version of this example had an example of negative sampling. For example of negative sampling, please refer to: Gluon NLP Sampled Block

Acknowledgements

Thanks to xlvector for the first Matrix Factorization example that provided the basis for these examples.

MovieLens data from GroupLens. Note: MovieLens 100K and 10M dataset are copy right to GroupLens Research Group at the University of Minnesota, and licensed under their usage license. For full text of the usage license, see ml-100k license and ml-10m license.