We introduce rrecsys, an open source extension package in R for rapid prototyping and intuitive assessment of recommender system algorithms. As the only currently available R package for recommender algorithms (recommenderlab) did not include popular algorithm implementations such as matrix factorization or One-class Collaborative Filtering algorithms we developed rrecsys as an easily accessible tool that can, for instance, be employed for interactive demonstrations when teaching. This package replicates state-of-the-art Collaborative Filtering algorithms for rating and binary data and we compare results with the Java-based LensKit implementation and recommederlab for the purpose of benchmarking the implementation. Therefore this work can also be seen as a contribution in the context of replication of algorithm implementations and reproduction of evaluation results.
R represents a popular choice in Data Analytics and
Machine Learning. The software has low setup cost and
contains a large selection of packages and functionalities to
enhance and prototype algorithms with compact code and
good visualization tools. Thus R represents a suitable
environment for exploring the eld of recommender systems.
We present and contribute a novel R package, rrecsys1, that
replicates several state-of-the-art recommender algorithms
for Likert scaled as well as binary rating values. Up to now
there is only one package addressing recommender systems,
recommenderlab 2, which lacks implementation of popular
algorithms and we benchmark results in Section 3.
This work can be seen as a contribution towards the
reproducibility of algorithms and results. Although this concept
of reproducibility of experimental results is a fundamental
prerequisite for scienti c research it is many times not given
for granted in the recommender systems eld. For instance
Said et al. [
rrecsys has a modular structure as well as includes expansion capabilities. The core of the package includes the implementation of several popular algorithms such as: Most Popular, Global Average, Item Average, User Average, Item Based K-Nearest Neighbors, Simon Funk's SVD, Weighted Alternated Least Squares and Bayesian Personalized Ranking. The package's evaluation module is based on k-fold cross-validation method. A strati ed random selection procedure is applied when dividing the rated items of each user into k folds such that each user is uniformly represented in each fold. Based on the task (rating prediction or recommendation) the following metrics are computed: mean absolute error(MAE), root mean square error(RMSE), Precision, Recall, F1, True and False Positives, True and False Negatives, normalized discounted cumulative gain (NDCG), rank score, area under the ROC curve (AUC) and catalog coverage. RMSE and MAE metrics are computed according to their two variants, user-based vs. global. 3.
In this section we introduce an executable script in R for running some of the functionalities of rrecsys in order to demonstrate its intuitive use. # Install and load: install.packages("rrecsys") library(rrecsys) # ML Latest is loaded on the package. data("mlLatest100k") # Define a rating matrix and explore it. mlLatest <- defineData(mlLatest100k,
minimum = .5, maximum = 5, halfStar = TRUE) sparsity(mlLatest); numRatings(mlLatest) rowRatings(mlLatest); colRatings(mlLatest) smallMlLatest <- mlLatest[rowRatings(mlLatest) >= 200, colRatings(mlLatest) > 10] # Setting up the number of iterations for FunkSVD. setStoppingCriteria(nrLoops = 50) # Training a model using FunkSVD. svd10 <- rrecsys(smallMlLatest, "FunkSVD", k = 10, lambda = 0.001, gamma = 0.0015) # Using the trained model to predict and recommend. p <- predict(svd10) r <- recommend(svd10, topN = 10) # Instantiate an evaluation model. model <- evalModel(smallMlLatest, folds = 5) # Using the above model to evaluate predictions. evalPred(model, "IBKNN", neigh = 10) # Using the same model to evaluate recommendations. evalRec(model, "globalAverage", topN = 10, goodRating = 3)
In Table 1 we report results from benchmarking the
rrecsys implementation with the popular Lenskit [
Baseline Alg.
BPR(20 features, 20 iteration) BPR(40 features, 20 iteration) wALS(20 features, 20 iteration) wALS(40 features, 20 iteration) 0:85 0:8 10 50 100 150 200
GB RAM.We report execution times for a single prediction task and the full evaluation steps in Table 2 and 3.
This poster contributed a recently released package for prototyping and interactively demonstrating recommendation algorithms in R. It comes with a nice range of implemented standard CF algorithms. Reported results demonstrate that it reproduces results of the Java-based Lenskit toolkit. Thus it remains to hope that this e ort will be of use for the eld of recommender systems and the large R user community.