Latent Variable Session-Based Recommendation

Stephen Bonner, David Rohde

Keywords: session based recommendation, latent variable, EM algorithm, Bouchard bound

TL;DR: Fast variational approximations for approximating a user state and learning product embeddings

Abstract: We present a probabilistic framework for session based recommendation. A latent variable for the user state is updated as the user views more items and we learn more about their interests. We provide computational solutions using both the re-parameterization trick and using the Bouchard bound for the softmax function, we further explore employing a variational auto-encoder and a variational Expectation-Maximization algorithm for tightening the variational bound. Finally we show that the Bouchard bound causes the denominator of the softmax to decompose into a sum enabling fast noisy gradients of the bound giving a fully probabilistic algorithm reminiscent of word2vec and a fast online EM algorithm.