recommender's Introduction

recommender

The recommender package provides collaborative filter based product recommendations.

Installation

The newest development release can be installed from GitHub:

# install.packages('devtools')
devtools::install_github("byapparov/recommender")

Similar products model

To create a model for similar products recommender you will need a history of user-to-product interactions in a data.frame where first collumn identifies the user and second column identifies a product, e.g.:

  user.hits <- data.table(
    users =    c("u1", "u2", "u1", "u3", "u2", "u1"),
    products = c("p1", "p2", "p3", "p2", "p3", "p4")
  )
  model <- similarityRecommender(user.hits)

Product interaction history columns are matched according to order:

User Identifier (any name)
Product Identifier (any name)

Any other columns in the table will be ignored.

Item-to-item recommendations

recommendComplimentaryProducts() function provides "others also viewed" type of recommendations.

 # table of products which will be linked to recommendations
 products <- data.table(sku = c("a", "b", "c", "d"),
                         type = c("p1", "p2", "p3", "p1"))

 # Get 5 most similar products for each product in the `products` table
 product.affinity <- recommendComplimentaryProducts(model, products, limit = 5)

User-to-item recommendations

recommendSimilarProducts() function provides "similar products" recommendations based on the user-item interactions data.

  # product interactions stream of new users
  page.views <- data.table(
    user = c("u1", "u1", "u2", "u3", "u3", "u3"),
    sku = c("a", "b",   "c",  "a",  "a", "d")
  )
  
  # `groups` in the filter limit number of recommended products
  # from the same group to one. this can be useful in cases where 
  # distance between recommended items should be increased
  # 
  # `values` in the filter limit number of items returned per user
  groups <- c("a" = "p1", "b" = "p2", "c" = "p3", "d" = "p1")
  filter <- makeRecommendationsFilter(groups, values = 1)
  
  # make user-to-item recommendations table
  res <- recommendSimilarProducts(  
    test.sim.model, 
    page.views, 
    exclude.same = T, 
    filter = filter
  )

recommender's People

Contributors

Watchers

recommender's Issues

Investigate if recommenderlab provides correct results

library(recommenderlab)
data <- readRDS("hits.rds")
data$hit <- 1L
m <- dcast(data, visitor.id ~ sku, mean, value.var = "hit", fill = 0L)
m <- data.matrix(m)
rm <- as(m, "realRatingMatrix")
m <- binarize(m, minRating = 1)

Recommend Projects