recommenderlab

You can install the stable version of recommenderlab from CRAN, or the current development version from r-universe.

# Stable CRAN version
install.packages("recommenderlab")

# Current development version
install.packages("recommenderlab",
    repos = c("https://mhahsler.r-universe.dev",
              "https://cloud.r-project.org/"))

The MovieLens dataset is provided for research purposes under specific conditions:

1. No Endorsement: You may not imply endorsement from the University of Minnesota or the GroupLens Research Group.
2. Acknowledgment: You must acknowledge the use of the dataset in any resulting publications and provide a copy of those publications to the researchers.
3. No Redistribution: You may not redistribute the data without separate permission.
4. Non-Commercial: You may not use this information for commercial or revenue-bearing purposes without prior permission from a faculty member of the GroupLens Research Project at the University of Minnesota.

recommenderlab is an R package that provides research infrastructure for developing and evaluating collaborative filtering recommender algorithms. It supports both rating data (e.g., 1-5 stars) and unary (0-1) datasets. Key features include:

• Sparse representation for user-item matrices.
• A variety of popular recommendation algorithms.
• Support for top-N recommendations.
• Robust cross-validation and evaluation frameworks.

To use the package, load the library and use the Recommender() function to train a model on a dataset (such as a realRatingMatrix). You can then use predict() to generate recommendations for new users.

Note: When predicting, you can specify n to get the top-N recommendations.

library("recommenderlab")
data("MovieLense")

# Prepare data (e.g., users with > 100 ratings)
MovieLense100 <- MovieLense[rowCounts(MovieLense) > 100, ]

# Train a user-based collaborative filtering (UBCF) model
train <- MovieLense100[1:300]
rec <- Recommender(train, method = "UBCF")

# Predict top-5 recommendations for users 301 and 302
pre <- predict(rec, MovieLense100[301:302], n = 5)

# Convert predictions to a list for easier viewing
as(pre, "list")

The users.dat file contains demographic information for users. Each line follows this format:

UserID::Gender::Age::Occupation::Zip-code

• UserID: Integer.
• Gender: M for male, F for female.
• Age: Represented by integer ranges:
  • 1: Under 18
  • 18: 18-24
  • 25: 25-34
  • 35: 35-44
  • 45: 45-49
  • 50: 50-55
  • 56: 56+
• Occupation: Integer mapping (e.g., 1 for academic/educator, 12 for programmer, 20 for writer).
• Zip-code: User-provided postal code.

The ratings.dat file contains user ratings for movies. Each line follows this format:

UserID::MovieID::Rating::Timestamp

• UserID: Integer between 1 and 6040.
• MovieID: Integer between 0 and 3592.
• Rating: A whole-star rating on a 5-star scale.
• Timestamp: Seconds since the epoch (as returned by time(2)).

The framework provides infrastructure for evaluating algorithms using several protocols and measures.

Evaluation Protocols

• Train/test split
• Cross-validation
• Repeated bootstrap sampling
• Given-n and all-but-x protocols

Evaluation Measures

• Rating errors: MSE, RMSE, MAE
• Top-N recommendations: TPR/FPR (ROC), precision, and recall

The movies.dat file contains movie metadata. Each line follows this format:

MovieID::Title::Genres

• MovieID: Integer.
• Title: Movie title (includes release year, matching IMDB).
• Genres: Pipe-separated (|) list of genres (e.g., Action|Adventure).

Supported genres include: Action, Adventure, Animation, Children's, Comedy, Crime, Documentary, Drama, Fantasy, Film-Noir, Horror, Musical, Mystery, Romance, Sci-Fi, Thriller, War, and Western.

The recommenderlab framework supports a wide variety of collaborative filtering and baseline algorithms:

• UBCF: User-based collaborative filtering
• IBCF: Item-based collaborative filtering
• SVD: SVD with column-mean imputation
• SVDF: Funk SVD
• ALS: Alternating Least Squares
• LIBMF: Matrix factorization with LIBMF
• AR: Association rule-based recommender
• POPULAR: Popular items
• RANDOM: Randomly chosen items for comparison
• RERECOMMEND: Re-recommend liked items
• HybridRecommender: Hybrid recommendations

You can compare multiple algorithms using an evaluationScheme. This involves defining a scheme (e.g., 10-fold cross-validation), specifying the algorithms and their parameters, and then calling evaluate().

# Define the evaluation scheme
scheme <- evaluationScheme(MovieLense100, method = "cross-validation", k = 10, 
  given = -5, goodRating = 4)

# Define algorithms to compare
algorithms <- list(
  "random items" = list(name = "RANDOM", param = NULL),
  "popular items" = list(name = "POPULAR", param = NULL),
  "user-based CF" = list(name = "UBCF", param = list(nn = 3)),
  "item-based CF" = list(name = "IBCF", param = list(k = 100))
)

# Run evaluation for top-N lists of different lengths
results <- evaluate(scheme, algorithms, type = "topNList",
  n = c(1, 3, 5, 10), progress = FALSE)

# Plot the results
plot(results, annotate = 2, legend = "topleft")

The package provides a framework for evaluating algorithms using different protocols and metrics.

Evaluation Protocols

• Train/test split
• Cross-validation
• Repeated bootstrap sampling

Protocols can be configured using given-n and all-but-x methods.

Evaluation Measures

• Rating errors: MSE, RMSE, MAE
• Top-N recommendations: TPR/FPR (ROC), precision, and recall