Matchbox: large scale online bayesian recommendations

By David H. Stern, Ralf Herbrich, Thore Graepel. 2009

277 citations recorded by the source corpus.

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

Ups and Downs: Modeling the Visual Evolution of Fashion Trends with One-Class Collaborative Filtering (2016, background): “Following this thread, there are also subsequent works that build probabilistic models (e.g. [29, 33]) to address the same task.”
Fixed-Form Variational Posterior Approximation through Stochastic Linear Regression (2012, background): “Although this is often desirable (see e.g. Stern et al., 2009), there is no guarantee that the recovered local minimum of the KL-divergence is then also a global minimum.”
Approximate Gaussian Integration using Expectation Propagation (2011, unknown): “pectation Propagation without empirical evaluation. However, EP has been shown to provide excellent approximations in many applications (Minka, 2001a; Kuss and Rasmussen, 2005; Herbrich et al., 2007; Stern et al., 2009), often outperforming competing methods. These successes have deservedly establis”
Strathprints Institutional Repository Bayesian Latent Variable Models for Collaborative Item Rating Prediction. In: Acm Cikm, 2011-03-31. Bayesian Latent Variable Models for Collaborative Item Rating Prediction (methodology): “[17] instead use Expectation Propagation and Variational Message Passing to learn a model using both ratings data and content.”
Improving pairwise learning for item recommendation from implicit feedback (2014, background): “The popular matrix factorization model (e.g. [16]), has been extended for many recommender scenarios, e.g. using implicit information [10, 18], time [11], neighborhood information [10] or attributes [20].”
Collaborative Filtering beyond the User-Item Matrix (2014, background): “In the most basic case, interactive recommendation is studied as a case of the problem of online CF [Liu et al. 2010; Stern et al. 2009].”
Refining Recency Search Results with User Click Feedback (2011, background): “As a similar, independent thread of work, [28] also considered the large-scale personalized recommendation problem, but imposed a Bayesian framework, which is diﬀerent from our work.”
Scalable Recommendation with Poisson Factorization (2013, background): “) MF has been extended in many ways to implement modern recommendation systems [7, 21, 27, 31].”
Collaborative Gaussian Processes for Preference Learning (2012, methodology): “This approach is similar to dimensionality reduction methods that are commonly used for addressing collaborative filtering problems [19, 16].”
Bayesian Matrix Factorization with Side Information and Dirichlet Process Mixtures (2010, methodology): “The Matchbox Bayesian recommendation system (Stern, Herbrich, and Graepel 2009) is another bilinear model featuring feedback and dynamics models and a similar mechanism to incorporate known information, with expectation propagation as the inference algorithm.”
Alors: An algorithm recommender system (2017, methodology): “…overall runtime less than one minute on a standard PC. Experiments on artiﬁcial data show the robustness of Alors compared to the Matchbox system (Stern et al., 2009, 2010), with respect to the initial representation of the problem instances; • The latent representation extracted by Alors…”
Recurrent Recommender Networks (2017, background): “Generative models for personalized ratings have recently become popular, due to impressive and robust results (Mnih & Salakhutdinov, 2007; Salakhutdinov & Mnih, 2008; Stern et al., 2009; Beutel et al., 2015).”
Cold-start Active Learning with Robust Ordinal Matrix Factorization (2014, methodology): “However, it is well known that for factors corresponding to the matrix factorizations, EP provides poor approximations (Stern et al., 2009), so for these we use VB. Implementational details are in the supplementary material.”
Probabilistic Matrix Factorization with Non-random Missing Data (2014, background): “…they can be robust to overﬁtting and come with automatic estimates of uncertainty (Mnih & Salakhutdinov, 2007); ii) they can be adapted to different data types, such as continuous, binary or ordinal data (Stern et al., 2009); and iii) they can theoretically handle missing data in a formal manner.”