Recommendation from Raw Data with Adaptive Compound Poisson Factorization

Abstract

Count data are often used in recommender systems: they are widespread (song play counts, product purchases, clicks on web pages) and can reveal user preference without any explicit rating from the user. Such data are known to be sparse, over-dispersed and bursty, which makes their direct use in recommender systems challenging, often leading to pre-processing steps such as binarization. The aim of this paper is to build recommender systems from these raw data, by means of the recently proposed compound Poisson Factorization (cPF). The paper contributions are three-fold: we present a unified framework for discrete data (dcPF), leading to an adaptive and scalable algorithm; we show that our framework achieves a trade-off between Poisson Factorization (PF) applied to raw and binarized data; we study four specific instances that are relevant to recommendation and exhibit new links with combinatorics. Experiments with three different datasets show that dcPF is able to effectively adjust to over-dispersion, leading to better recommendation scores when compared with PF on either raw or binarized data.

Hierarchical Negative Binomial Factorization for Recommender Systems on Implicit Feedback

“…Gouvert et al (Gouvert, Oberlin, and Févotte 2018) assume data to be NB and propose negative binomial factorization (NBF) with the high computational cost. Gouvert et al (Gouvert, Oberlin, and Févotte 2019) also propose a compound PF with a dispersion model to represent user behavior in listening sessions. All the previous compound PF-based methods merely utilize a single-layered dispersion model to be accessible to various distribution assumptions of dispersion for generality, which is doubted to adaptive to various user behaviors.…”

“…By contrast, the number of times a user neglected an exposed item is called failure exposure count (FEC). To model FEC, prior works (Gouvert, Oberlin, and Févotte 2018;Zhou 2018;Gouvert, Oberlin, and Févotte 2019) assume an entry in a utility matrix to be a negative binomial distribution (NB). However, in those NB-based models, the structure for modeling FEC is single-layered to avoid…”