Matrix Factorization based recommender system in Go. Because facts are more important than ever.
This project provides a vectormodel package that can be used to serve real time recommendations. First of all, you will need to train a model to get document embeddings or latent factors. I highly recommend the implicit library for that. Once you have the documents as a map of int ids to arrays of float64, you can create the vector model by calling:
model, err := NewVectorModel(documents map[int][]float64, confidence, regularization float64)
And to generate recommendations call .Recommend with a set of items the user has seen:
recs := model.Recommend(seenDocs *map[int]bool, n int)
Note that user vectors are not required. Matter of fact, you can use this to recommend documents to users that were not in the training set. The recommendations will be computed very efficiently (probably <1ms, depends on your model size) in real time.
Check out the demo for a complete example that recommends GitHub repositories.
Demo source code is available here: https://github.com/jbochi/github-recs