Collaborative prediction and ranking with non-random missing data

By Benjamin M. Marlin, Richard S. Zemel. 2009

389 citations recorded by the source corpus.

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

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 Marketing

Citations in the corpus

On the challenges of learning with inference networks on sparse, high-dimensional data (2017, methodology): “We study the top-N recommendation performance of NFA under strong generalization (Marlin & Zemel, 2009).”
How algorithmic confounding in recommendation systems increases homogeneity and decreases utility (2017, background): “Other work also considers similar notions of missingness in rating data [43, 62].”
Recommendations as Treatments: Debiasing Learning and Evaluation (2016, background): “This creates a widely-recognized challenge for evaluating recommender systems (Marlin & Zemel, 2009; Myttenaere et al., 2014).”
Scalable Recommendation with Poisson Factorization (2013, methodology): “…figure, we illustrate three distributions over user activity: the observed distribution (squares), the distribution from a 1Specifically, by classical matrix factorization we mean L2 regularized matrix factorization with bias terms for users and items, fit using stochastic gradient descent [21].”
Transparent, Scrutable and Explainable User Models for Personalized Recommendation (2019, background): “In practice, these are biased towards items users like or know [29, 45].”
Cold-start Active Learning with Robust Ordinal Matrix Factorization (2014, methodology): “We compare our model for heteroskedastic ordinal matrix factorization (HOMF) against the following methods: i) the homoskedastic model with an ordinal likelihood in Paquet et al. (2012) (Paquet); ii) a method for robust Bayesian matrix factorization (RBMF) based on a Gaussian likelihood which includ”
Ranking with non-random missing ratings: influence of popularity and positivity on evaluation metrics (2012, methodology): “The bias involved by users choosing the items they rate, and its influence on model fitting has been cast in light in [8] (with, in addition, the release of the Yahoo! dataset we use in this paper), and further studied in [9].”
Probabilistic Matrix Factorization with Non-random Missing Data (2014, methodology): “Marlin & Zemel (2009) have proposed a method for modeling MNAR rating data in collaborative ﬁltering.”
Recommender Systems (2016, neutral): “This property of rating distributions is also known as Missing Not At Random (MNAR), or selection bias [402, 565].”
Graphical Models for Inference with Missing Data (2013, background): “Exceptions are recent work on collaborative filtering and recommender systems which develops probabilistic models that explicitly incorporate missing data mechanism [16, 14, 15].”
In2Rec: Influence-based Interpretable Recommendation (2019, methodology): “Marlin et al. [19] integrated the non-random missing data mechanism into multinomial mixture model.”
Collaborative Filtering Recommender Systems (2011, background): “For instance, researchers have shown that datasets can be effectively statistically unbiased for some popularity biases [88].”
Training and testing of recommender systems on data missing not at random (2010, background): “This kind of data is inexpensive to collect, and typically used for training and testing recommender systems (prior to conducting more costly (cid:12)eld studies).”
Top-N Recommendation with Missing Implicit Feedback (2015, methodology): “[5, 2] have studied the MNAR assumption in terms of model fitting with different missing data models but the evaluations do not take the missing data models into account.”
Beyond Parity: Fairness Objectives for Collaborative Filtering (2017, background): “Unfortunately, researchers have shown that sampled ratings have markedly different properties from the users’ true preferences [21, 22].”
Partially Observable Markov Decision Process for Recommender Systems (2016, background): “On one hand, the researchers proposed to augment the training data by generating negative samples [15, 11].”