An Empirical Analysis of Design Choices in Neighborhood-Based Collaborative Filtering Algorithms

By Jonathan L. Herlocker, J. Konstan, John Riedl. Information retrieval (Boston), 2002

733 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Information Systems, Signal Processing

View Paper: https://www.semanticscholar.org/paper/eafcbaa42e262ffb9be5095aa7e7cde91d7a2828

How researchers see this paper

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

In Artificial Intelligence

In Information Systems

In Signal Processing

Citations in the corpus

Factorization Meets the Item Embedding: Regularizing Matrix Factorization with Item Co-occurrence (2016, background): “Previous work shows that the performance of item-based collaborative filtering is highly sensitive to the choice of similarity metric and data normalization [8].”
Collaborative filtering recommender systems (2013, background): “First user’s list of ratings is turned into a list of ranks, where the user’s highest rating takes the rank of 1 and tied ratings take the average of the ranks for their spot (Herlocker et al ., 2002).”
Recommender Systems (2016, unknown): “The specific advantage of these distance measures is not quite clear because conflicting results have been presented in the literature [247, 258].”
Context Trees: Crowdsourcing Global Understanding from Local Views (2014, methodology): “In addition, workers will use different ranges of the rating scale (Herlocker, Konstan, and Riedl 2002).”
A new collaborative filtering metric that improves the behavior of recommender systems (2010, background): “Despite the deficiencies of Pearson correlation, this similarity measure presents the best prediction and recommendation results in CF-based RS [15,16,31,7,35], furthermore, it is the most commonly used, and therefore, any alternative metric proposed must improve its results.”
A Comprehensive Survey of Neighborhood-based Recommendation Methods (2011, methodology): “In [29, 28], it was found that using γ ≥ 25 could significantly improve the accuracy of the predicted ratings, and that a value of 50 for γ gave the best results.”
Challenging the Long Tail Recommendation (2012, methodology): “Similarly, the recommendation results from the classic collaborative ﬁlter-ing(e.g. [12, 20]) are always local popular and obvious.”
Recommender systems survey (2013, background): “[93] provides a set of guidelines for designing neighborhood-based prediction systems.”
Evaluating Recommendation Systems (2011, background): “As recommender systems are expected in many cases to provide rapid recommendations online, it is also important to measure how fast does the system provides recommendations [23, 50].”
Tag recommendations based on tensor dimensionality reduction (2008, background): “A more severe disadvantage is presented from regular recommender systems, like Collaborative Filtering (CF), [2, 6, 7, 11], which are applied only to 2-dimensional data (e.”
Multi-Interest Network with Dynamic Routing for Recommendation at Tmall (2019, methodology): “e network for capturing more features. User Representation. Representing users as vectors is commonly used in RS. Traditional methods assembles user preference as vectors composed of interested items [4, 12, 22], keywords [5, 8] and topics [29]. As the emergence of distributed representation learnin”
Evaluating collaborative filtering recommender systems (2004, unknown): “…example of how this could be significant, consider the following situation that could occur when using the nearest neighbor algorithm described in [Herlocker et al. 2002]: when only one user in the dataset has rated an eclectic item I, then the prediction for item I for all users will be equal to…”
A collaborative filtering algorithm and evaluation metric that accurately model the user experience (2004, background): “In nearest neighbor algorithms, it has been shown that normalizing a neighbor’s ratings with respect to their average rating provides a substantial increase in algorithm accuracy [3].”
Analysis of the Mean Absolute Error (MAE) and the Root Mean Square Error (RMSE) in Assessing Rounding Model (2018, methodology): “We compare the following algorithms with Basic Rounding approach: user-based kNN (using Pearson similarity and the neighborhood size of 30 which is suggested as the optimal value in literature [2]), Slope One [8], and Classical MF [9].”
Recommender Systems in Technology Enhanced Learning (2011, methodology): “…is a common evaluation practice in systems examined for other domains (e.g. Breese et al. 1998; Deshpande and Karypis 2004; Papagelis et al. 2005; Herlocker et al. 2002), which indicate that careful testing and parameterisation has to be carried out before a re-commender system is finally…”
Collaborative Filtering Recommender Systems (2011, methodology): “A straightforward method of measuring the recommendation quality is to measure the mean absolute error (MAE) [22, 58, 59, 112, 137], sometimes also called absolute deviation.”