GaussianProductAttributes: Density-Based Distributed Representations for Products

Multivariate Gaussian probability distributions have been used as distributed representations for text. In comparison with traditional vector representations, these density-based representations are able to model uncertainty, inclusion and entailment. We present a model to learn such representations for products based on a public e-commerce dataset. We qualitatively analyse the properties of the proposed model and how the learned representations capture semantic relatedness, similarity and entailment between products and text.