neural networks. We then apply our learnt product and user representations on two recommendation frameworks based on matrix factorization and link prediction to enhance the recommendation task. Experimental results on four datasets constructed from...