I am using Recommenderlab in R to build a recommendation system to provide craft-beer suggestions to new users.
However, upon running the model, I am receiving the same predictions per user for a majority of the training dataset, or receiving 'character(0)' as the output. How can I receive the predictions that are associated with each user and not duplicated?
The dataset I'm using can be found here: https://www.kaggle.com/rdoume/beerreviews/version/1
I have tried converting the data frame directly into a matrix, then into a realRatingMatrix.
In order to receive any recommendations, I need to use the 'dcast' function from the data.table library before converting the data frame into a matrix.
I have also tried removing the first column from the matrix to drop the user ids.
One thing to note is that when the data is sampled, there can be a few rows where the 'reviewer' is blank, but the rating and beer id is there.
library(dplyr)
library(tidyverse)
library(recommenderlab)
library(reshape2)
library(data.table)
beer <- read.csv('beer.csv', stringsAsFactors = FALSE)
#Take sample of data(1000)
beer_sample <- sample_n(beer, 1000)
#Select relevant columns & rename
beer_ratings <- select(beer_sample, reviewer = review_profilename, beerId = beer_beerid, rating = review_overall)
#Add unique id for reviewers
beer_ratings$userId <- group_indices_(beer_ratings, .dots = 'reviewer')
#Create ratings matrix
rating_matrix <- dcast(beer_ratings, userId ~ beerId, value.var = 'rating')
rating_matrix <- as.matrix(rating_matrix)
rating_matrix <- as(rating_matrix, 'realRatingMatrix')
#UBCF Model
recommender_model <- Recommender(rating_matrix, method = 'UBCF', param=list(method='Cosine',nn=10))
#Predict top 5 beers for first 10 users
recom <- predict(recommender_model, rating_matrix[1:10], n=5)
#Return top recommendations as a list
recom_list<- as(recom,'list')
recom_list
The above code will result in:
[[1]]
[1] "48542" "2042" "6" "10" "19"
[[2]]
[1] "10277" "2042" "6" "10" "19"
[[3]]
[1] "10277" "48542" "6" "10" "19"
[[4]]
[1] "10277" "48542" "2042" "6" "10"
[[5]]
[1] "10277" "48542" "2042" "6" "10"
[[6]]
[1] "10277" "48542" "2042" "6" "10"
Converting the data frame to a matrix then realRatingMatrix without casting first into a table results in the user's recommendation as:
`886093`
`character(0)`
Using the 'dcast' function first then converting the data frame into a matrix and removing the first column, then into a realRatingMatrix returns the same predictions for almost every user:
[[1]]
[1] "6" "7" "10" "12" "19"
[[2]]
[1] "6" "7" "10" "12" "19"
[[3]]
[1] "6" "7" "10" "12" "19"
Any help is greatly appreciated.