Tensorized Determinantal Point Processes for Recommendation

Abstract

Interest in determinantal point processes (DPPs) is increasing in machine learning due to their ability to provide an elegant parametric model over combinatorial sets. In particular, the number of required parameters in a DPP grows only quadratically with the size of the ground set (e.g., item catalog), while the number of possible sets of items grows exponentially. Recent work has shown that DPPs can be effective models for product recommendation and basket completion tasks, since they are able to account for both the diversity and quality of items within a set. We present an enhanced DPP model that is specialized for the task of basket completion, the tensorized DPP. We leverage ideas from tensor factorization in order to customize the model for the next-item basket completion task, where the next item is captured in an extra dimension of the model. We evaluate our model on several real-world datasets, and find that the tensorized DPP provides significantly better predictive quality in several settings than a number of state-of-the art models.