Ulrich Paquet

One-class collaborative filtering with random graphs

The bane of one-class collaborative filtering is interpreting and modelling the latent signal from the missing class. In this paper we present a novel Bayesian generative model for implicit collaborative filtering. It forms a core component of the Xbox Live architecture, and unlike previous approaches, delineates the odds of a user disliking an item from simply being unaware of it. The latent signal is treated as an unobserved random graph connecting users with items they might have encountered. We demonstrate how large-scale distributed learning can be achieved through a combination of stochastic gradient descent and mean field variational inference over random graph samples. A fine-grained comparison is done against a state of the art baseline on real world data.

Speeding up the Xbox recommender system using a euclidean transformation for inner-product spaces

A prominent approach in collaborative filtering based recommender systems is using dimensionality reduction (matrix factorization) techniques to map users and items into low-dimensional vectors. In such systems, a higher inner product between a user vector and an item vector indicates that the item better suits the users preference. Traditionally, retrieving the most suitable items is done by scoring and sorting all items. Real world online recommender systems must adhere to strict response-time constraints, so when the number of items is large, scoring all items is intractable. We propose a novel order preserving transformation, mapping the maximum inner product search problem to Euclidean space nearest neighbor search problem. Utilizing this transformation, we study the efficiency of several (approximate) nearest neighbor data structures. Our final solution is based on a novel use of the PCA-Tree data structure in which results are augmented using paths one hamming distance away from the query (neighborhood boosting). The end result is a system which allows approximate matches (items with relatively high inner product, but not necessarily the highest one). We evaluate our techniques on two large-scale recommendation datasets, Xbox Movies and Yahoo~Music, and show that this technique allows trading off a slight degradation in the recommendation quality for a significant improvement in the retrieval time.

A hierarchical model for ordinal matrix factorization

This paper proposes a hierarchical probabilistic model for ordinal matrix factorization. Unlike previous approaches, we model the ordinal nature of the data and take a principled approach to incorporating priors for the hidden variables. Two algorithms are presented for inference, one based on Gibbs sampling and one based on variational Bayes. Importantly, these algorithms may be implemented in the factorization of very large matrices with missing entries.The model is evaluated on a collaborative filtering task, where users have rated a collection of movies and the system is asked to predict their ratings for other movies. The Netflix data set is used for evaluation, which consists of around 100 million ratings. Using root mean-squared error (RMSE) as an evaluation metric, results show that the suggested model outperforms alternative factorization techniques. Results also show how Gibbs sampling outperforms variational Bayes on this task, despite the large number of ratings and model parameters. Matlab implementations of the proposed algorithms are available from cogsys.imm.dtu.dk/ordinalmatrixfactorization.

Xbox movies recommendations: variational bayes matrix factorization with embedded feature selection

We present a matrix factorization model inspired by challenges we encountered while working on the Xbox movies recommendation system. The item catalog in a recommender system is typically equipped with meta-data features in the form of labels. However, only part of these features are informative or useful with regard to collaborative filtering. By incorporating a novel sparsity prior on feature parameters, the model automatically discerns and utilizes informative features while simultaneously pruning non-informative features. The model is designed for binary feedback, which is common in many real-world systems where numeric rating data is scarce or non-existent. However, the overall framework is applicable to any likelihood function. Model parameters are estimated with a Variational Bayes inference algorithm, which is robust to over-fitting and does not require cross-validation and fine tuning of regularization coefficients. The efficacy of our method is illustrated on a sample from the Xbox movies dataset as well as on the publicly available MovieLens dataset. In both cases, the proposed solution provides superior predictive accuracy, especially for long-tail items. We then demonstrate the feature selection capabilities and compare against the common case of simple Gaussian priors. Finally, we show that even without features, our model performs better than a baseline model trained with the popular stochastic gradient descent approach.

The Xbox recommender system

A recent addition to Microsofts Xbox Live Marketplace is a recommender system which allows users to explore both movies and games in a personalized context. The system largely relies on implicit feedback, and runs on a large scale, serving tens of millions of daily users. We describe the system design, and review the core recommendation algorithm.

A large-scale exploration of group viewing patterns

We present a large-scale study of television viewing habits, focusing on how individuals adapt their preferences when consuming content with others. While there has been a great deal of research on modeling individual preferences, there has been considerably less work studying the preferences of groups, due mostly to the difficulty of collecting group data. In contrast to most past work that has relied either on small-scale surveys, prototypes, or a relatively limited amount of group preference data, we explore more than 4 million logged household views paired with individual-level demographic and co-viewing information. Our analysis reveals how engagement in group viewing varies by viewer and content type, and how viewing patterns shift across various group contexts. Furthermore, we leverage this large-scale dataset to directly estimate how individual preferences are combined in group settings, finding subtle deviations from traditional models of preference aggregation. We present a simple model which captures these effects and discuss the impact of these findings on the design of group recommendation systems.

Bayesian Low-Rank Determinantal Point Processes

Determinantal point processes (DPPs) are an emerging model for encoding probabilities over subsets, such as shopping baskets, selected from a ground set, such as an item catalog. They have recently proved to be appealing models for a number of machine learning tasks, including product recommendation. DPPs are parametrized by a positive semi-definite kernel matrix. Prior work has shown that using a low-rank factorization of this kernel provides scalability improvements that open the door to training on large-scale datasets and computing online recommendations, both of which are infeasible with standard DPP models that use a full-rank kernel. A low-rank DPP model can be trained using an optimization-based method, such as stochastic gradient ascent, to find a point estimate of the kernel parameters, which can be performed efficiently on large-scale datasets. However, this approach requires careful tuning of regularization parameters to prevent overfitting and provide good predictive performance, which can be computationally expensive. In this paper we present a Bayesian method for learning a low-rank factorization of this kernel, which provides automatic control of regularization. We show that our Bayesian low-rank DPP model can be trained efficiently using stochastic gradient Hamiltonian Monte Carlo (SGHMC). Our Bayesian model generally provides better predictive performance on several real-world product recommendation datasets than optimization-based low-rank DPP models trained using stochastic gradient ascent, and better performance than several state-of-the art recommendation methods in many cases.

Gaussian process modeling for image distortion correction in echo planar imaging

An enhanced method for correction of image distortion due to B0‐field inhomogeneities in echo planar imaging (EPI) is presented. The algorithm is based on the measurement of the point spread function (PSF) associated with each image voxel using a reference scan. The expected distortion map in the phase encode direction is then estimated using a nonparametric inference algorithm known as Gaussian process modeling. The algorithm is shown to be robust to the presence of regions of low signal‐to‐noise in the image and large inhomogeneities. Magn Reson Med, 2008.

Beyond Collaborative Filtering: The List Recommendation Problem

Most Collaborative Filtering (CF) algorithms are optimized using a dataset of isolated user-item tuples. However, in commercial applications recommended items are usually served as an ordered list of several items and not as isolated items. In this setting, inter-item interactions have an effect on the lists Click-Through Rate (CTR) that is unaccounted for using traditional CF approaches. Most CF approaches also ignore additional important factors like click propensity variation, item fatigue, etc. In this work, we introduce the list recommendation problem. We present useful insights gleaned from user behavior and consumption patterns from a large scale real world recommender system. We then propose a novel two-layered framework that builds upon existing CF algorithms to optimize a lists click probability. Our approach accounts for inter-item interactions as well as additional information such as item fatigue, trendiness patterns, contextual information etc. Finally, we evaluate our approach using a novel adaptation of Inverse Propensity Scoring (IPS) which facilitates off-policy estimation of our methods CTR and showcases its effectiveness in real-world settings.

Bayesian hierarchical ordinal regression

We present a Bayesian approach to ordinal regression. Our model is based on a hierarchical mixture of experts model and performs a soft partitioning of the input space into different ranks, such that the order of the ranks is preserved. Experimental results on benchmark data sets show a comparable performance to support vector machine and Gaussian process methods.