Empirical Analysis of Predictive Algorithms for Collaborative Filtering

By J. Breese, David E. Heckerman, Carl Myers Kadie. 1998

Collaborative filtering or recommender systems use a database about user preferences to predict additional topics or products a new user might like. In this paper we describe several algorithms designed for this task, including techniques based on correlation coefficients, vector-based similarity calculations, and statistical Bayesian methods. We compare the predictive accuracy of the various methods in a set of representative problem domains. We use two basic classes of evaluation metrics. The first characterizes accuracy over a set of individual predictions in terms of average absolute deviation. The second estimates the utility of a ranked list of suggested items. This metric uses an estimate of the probability that a user will see a recommendation in an ordered list. Experiments were run for datasets associated with 3 application areas, 4 experimental protocols, and the 2 evaluation metr rics for the various algorithms. Results indicate that for a wide range of conditions, Bayesian networks with decision trees at each node and correlation methods outperform Bayesian-clustering and vector-similarity methods. Between correlation and Bayesian networks, the preferred method depends on the nature of the dataset, nature of the application (ranked versus one-by-one presentation), and the availability of votes with which to make predictions. Other considerations include the size of database, speed of predictions, and learning time.

How researchers see this paper

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

Citations in the corpus

Geometric Matrix Completion with Recurrent Multi-Graph Neural Networks (2017, background): “Two major approach to recommender systems are collaborative (Breese et al., 1998) and content (Pazzani & Billsus, 2007) ﬁltering techniques.”
An MDP-Based Recommender System (2002, neutral): “We used e&¤A in our experiments in order to be consistent with the experiments of Breese et al. (1998) and Zimdars et al. (2001). The relative performance of the models was not sensitive to .”
SoRec: social recommendation using probabilistic matrix factorization (2008, methodology): “The most analyzed examples of memory-based collaborative filtering include userbased approaches [2, 7, 10, 25] and item-based approaches [4, 12, 19].”
The multiple multiplicative factor model for collaborative filtering (2004, methodology): “Note that this protocol is diﬁerent than the weak generalization protocol normally used for collaborative ﬂltering where methods are tested on held out training user ratings (Breese et al., 1998).”
Slope One Predictors for Online Rating-Based Collaborative Filtering (2007, methodology): “In computing MAE, we successively hide ratings one at a time from all evaluations in the test set while predicting the hidden rating, computing the average error we make in the prediction.”
Trust in recommender systems (2005, background): “Instead, collaborative filtering (sometimes called social filtering) relies on the availability of user profiles that capture the past ratings histories of users [3, 16].”
Context-aware recommender systems (2008, unknown): “The traditional two-dimensional (2D) neighborhood-based approach [21, 62] can be”
A recommender system based on collaborative filtering using ontology and dimensionality reduction techniques (2018, background): “Breese et al. (1998) further investigated the ole of clustering for the accuracy problem of recommender sysems.”
Identifying Opportunities for Valuable Encounters: Toward Context-Aware Social Matching Systems (2015, methodology): “Memory-based approaches [Breese et al. 1998; Linden et al. 2003; Resnick et al. 1994] use heuristics to make rating predictions based on the entire collection of items previously rated by users, whereas model-based CF approaches [Breese et al. 1998; Getoor and Sahami 1999; Hofmann 2003] use the…”
Making sense of recommendations (2019, background): “By contrast, user‐based collaborative filtering (“UBCF”) algorithms gather users from the dataset that have similar preferences on the sample ratings and make predictions by taking a weighted average of these users' ratings on the target item (Breese et al., 1998).”
Leveraging Missing Ratings to Improve Online Recommendation Systems (2006, methodology): “The popularity of these heuristic approaches stems from their ease of implementation, but they are often ad hoc and have been shown to be broadly inferior to model-based methods (Breese, Heckerman, and Kadie 1998).”
Improving the quality of predictions using textual information in online user reviews (2013, methodology): “This task of preference matching is carried out by recommendation systems [5].”
Deep reinforcement learning for page-wise recommendations (2018, methodology): “• CF [4]: Collaborative filtering is a method of making automatic predictions about the interests of a user by collecting preference information from many users, which is based on the hypothesis that people often get the best recommendations from someone with similar tastes to themselves.”
Mind the gaps: weighting the unknown in large-scale one-class collaborative filtering (2009, background): “Half-Life Utility (HLU) [4] estimates how likely a user will view/choose an item from a ranked list, which assumes that the user will view each consecutive item in the list with an exponential decay of possibility.”
Toward the next generation of recommender systems: a survey of the state-of-the-art and possible extensions (2005, methodology): “Both [15] and [11] compare their respective model-based approaches with standard memory-based approaches and report that, in some applications, model-based methods outperform memory-based approaches in terms of accuracy of recommendations.”