The BellKor Solution to the Netflix Grand Prize

By Yehuda Koren. 2009

531 citations recorded by the source corpus.

How researchers see this paper

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

In Recommender Systems

Citations in the corpus

No Spurious Local Minima in Nonconvex Low Rank Problems: A Unified Geometric Analysis (2017, methodology): “Such problems are studied extensively [Fazel, 2002, Rennie and Srebro, 2005, Candès and Recht, 2009] and has many applications in recommendation systems [Koren, 2009], see survey by Davenport and Romberg [2016].”
Model recommendation for action recognition (2012, methodology): “We initially explore the collaborative filtering approaches used by the BellKor team to (as a combined effort with two other groups) win the Netflix prize [3,12].”
Self-organization and missing values in SOM and GTM (2015, background): “, [12]) was an open competition to devise the best recommendation system to predict user ratings for films based on previous ratings.”
Recommender Systems (2016, unknown): “In fact, both winning entries in the Netflix Prize content, corresponding to Bellkor’s Pragmatic Chaos [311] and The Ensemble [704], used such an approach.”
A Survey of Graph Neural Networks for Recommender Systems: Challenges, Methods, and Directions (2021, background): “For example, Netflix’s movie recommendation competition [7] motivated many pioneer recommendation researches [7, 84].”
Neural Collaborative Filtering vs. Matrix Factorization Revisited (2020, background): “85 [21] – for Movielens 10M, the best models have about 0.”
Solving the apparent diversity-accuracy dilemma of recommender systems (2008, unknown): “It is interesting to note that while the Netflix Prize focused solely on accuracy, the winning entry in fact took a diversification approach, in this case based on tracking the changes in user opinions over time [32].”
Factorization Machines with libFM (2012, methodology): “The most successful approaches on Net.ix are based on matrix factorization (e.g., [Jahrer et al. 2010; Koren 2009a; Tak´ acs et al. 2009]).”
Ensemble contextual bandits for personalized recommendation (2014, methodology): “Ensemble Recommendation: Ensemble based algorithms have been well explored to improve the performance of prediction [21, 29], and are often preferred in recommendation competitions, such as the Netflix Prize contest [12, 26, 31] and KDD Cups [18, 34, 33].”
A Distributed, Architecture-Centric Approach to Computing Accurate Recommendations from Very Large and Sparse Datasets (2011, background): “claim with their extensive work on the Netflix dataset [13, 81].”
Protecting against evaluation overfitting in empirical reinforcement learning (2011, methodology): “For instance, the well-publicized Netflix competition used a single-environment methodology with holdout data, encouraging innovation by many teams over a multi-year period [25].”
Recommender Systems (2012, background): “combined team of BellKor [22], Pragmatic Theory [23] and BigChaos [24] (of course, it was not a simple combination but a sophisticated design), and each of them consists of many individual algorithms.”
Ensemble Methods as a Defense to Adversarial Perturbations Against Deep Neural Networks (2017, methodology): “These methods frequently win machine learning competitions, for example the Netflix prize [19].”
Identifying Platform Effects in Social Media Data (2016, methodology): “In the Netﬂix Prize data set, Koren (2009b), a member of the team that ultimately won the prize (Koren 2009a), points out a curious spike in the average ratings in early 2004.”