Scalable Recommendation with Poisson Factorization

By Prem Gopalan, Jake M. Hofman, David M. Blei. 2013

159 citations recorded by the source corpus.

We develop a Bayesian Poisson matrix factorization model for forming recommendations from sparse user behavior data. These data are large user/item matrices where each user has provided feedback on only a small subset of items, either explicitly (e.g., through star ratings) or implicitly (e.g., through views or purchases). In contrast to traditional matrix factorization approaches, Poisson factorization implicitly models each user's limited attention to consume items. Moreover, because of the mathematical form of the Poisson likelihood, the model needs only to explicitly consider the observed entries in the matrix, leading to both scalable computation and good predictive performance. We develop a variational inference algorithm for approximate posterior inference that scales up to massive data sets. This is an efficient algorithm that iterates over the observed entries and adjusts an approximate posterior over the user/item representations. We apply our method to large real-world user data containing users rating movies, users listening to songs, and users reading scientific papers. In all these settings, Bayesian Poisson factorization outperforms state-of-the-art matrix factorization methods.

Subfields: Information Systems, Management Science and Operations Research, Transportation

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Information Systems

In Management Science and Operations Research

In Transportation

Citations in the corpus

A Neural Autoregressive Approach to Collaborative Filtering (2016, methodology): “Among all these models, matrix factorization is most popular and successful, c.f. (Ko-ren et al., 2009; Salakhutdinov & Mnih, 2008; Mackey et al., 2011; Gopalan et al., 2013).”
Counterfactual inference for consumer choice across many product categories (2019, methodology): “From a machine learning perspective, we extend matrix factorization techniques developed by Gopalan et al. (2013) to focus on the case of shopping, which requires incorporating time-varying prices and demand shifts as well as an appropriate functional form.”
Deep Exponential Families (2014, methodology): “We note that the one layer sparse gamma DEF is equivalent to Poisson matrix factorization [9, 13] but our model is fully Bayesian and our variational distribution is collapsed.”
DropoutNet: Addressing Cold Start in Recommender Systems (2017, methodology): “Collaborative topic Poisson factorization (CTPF) [8] uses a similar interpolation architecture but replaces both LDA and WMF components with Poisson factorization [9].”
Content-based recommendations with Poisson factorization (2014, methodology): “[9], we add K latent variables z dv,k ∼ Poisson ( θ dk β vk ) , which are integers such that w dv = k z dv,k .”
Factorized Variational Autoencoders for Modeling Audience Reactions to Movies (2017, background): “Probabilistic versions have been proposed in [36, 41], where the data likelihood is typically Gaussian, but has been recently generalized to Poisson [12] for sparse count data.”
Bayesian Nonparametric Poisson Factorization for Recommendation Systems (2014, methodology): “For convenience, and similarly to previous works (Dunson & Herring, 2005; Zhou et al., 2012; Gopalan et al., 2013), we introduce for each user-item pair”
Feature selection and dimension reduction for single-cell RNA-Seq based on a multinomial model (2019, background): “Since in practice ni is large, the Poisson approximation to the multinomial may be reasonable [44, 45, 46, 47].”
Content-Aware Collaborative Music Recommendation Using Pre-trained Neural Networks (2015, methodology): “As shown in [4], the inference algorithm has complexity that scales linearly with the number of non-zero entries in the user-item matrix.”
Toward Implicit Sample Noise Modeling: Deviation-driven Matrix Factorization (2016, background): “The usage of exponential prior leads to sparse solu-tion (Gopalan, Hofman, and Blei 2013), which is favorable for its efﬁciency in computation and storage.”