Deep Learning Model - RBM(Restricted Boltzmann Machine) using Tensorflow for Products Recommendation

Product recommendations?are a must-have feature for all e-commerce websites, as they can drive sales, increase conversion rate and order value. According to Barilliance, “one-third of e-commerce revenues were generated from personalized?product recommendations. There are multiple algorithms used in past to perform product recommendations like- Content-Based Algorithms, Collaborative Filtering which comprised of User-User Collaborative Filtering, Item-based Collaborative filtering, and Market Basket Analysis. In the era of Deep Learning- there are multiple papers published on product recommendation using Deep learning models like seq2seq model(LSTM/GRU), Restricted Boltzmann Machine, Auto Encoders. I am going to use RBM(Restricted Boltzmann Machine) on the Movie Lens Data set.

RBM - RBMs are made out of input and hidden layers. You're trying to find a stochastic representation of the input. By sampling from the hidden layer, you can reproduce variants of samples encountered during training. The training of RBMs emerges through an alternate sampling of both layers, although backpropagation could still be used later to fine-tune the model. RBMs also seems more appropriate for recommender system applications as they can be used as means for collaborative filtering.

RBM are stochastic neural networks with two layers only: a layer of I visible units v, which is both designed for input and output, and a layer of J hidden units h. The number of visible units is the dimension of examples: I = M(Movies). The two layers are fully interconnected, but there is no connection within each layer. Neurons have a binary response. For detailed, mathematical understanding read this blog about RBM. RBM is an alternatives method for Product Recommendation

Please find below python code snippets to build RBM using Tensorflow library