Post-hoc explanations for complex model recommendations using simple methods

Many leading approaches for generating recommendations, such as matrix factorization and autoencoders, compute a complex model composed of latent variables. As such, explaining the recommendations generated by these models is a difficult task. In this paper, instead of attempting to explain the latent variables, we provide post-hoc explanations for why a recommended item may be appropriate for the user, by using a set of simple, easily explainable recommendation algorithms. When the output of the simple explainable recommender agrees with the complex model on a recommended item, we consider the explanation of the simple model to be applicable. We suggest both simple collaborative filtering and content based approaches for generating these explanations. We conduct a user study in the movie recommendation domain, showing that users accept our explanations, and react positively to simple and short explanations, even if they do not truly explain the mechanism leading to the generated recommendations.