Recommenderlab not working with highly sparse binary data?

Question

The data used is a ratings matrix generated from simple 0-1 yes/no click data based on whether or not a user visited a section of a website. This is implicit voting since if a user is interested in a section he will click it and thus rate it as a "1" and if not rate it as a "0". The ratings matrix thus has 32711 users in the rows and 285 site sections in the columns, after removing 9 that had no user visits.

I run:

library("recommenderlab")
ratingmatrix <- read.csv("ratingmatrix3.txt",",",header=TRUE, row.names=1 )
ratingmatrix <- as.matrix(ratingmatrix)
ratingmatrix <- as(ratingmatrix, "realRatingMatrix")

Rec1 <- Recommender(ratingmatrix[1:1000], method="POPULAR")
pred1 <- predict(Rec1, ratingmatrix[1001:1011], type="ratings")
as(pred1,"matrix")[,1:10]

I only get a prediction matrix with NAs implying that no predictions were made. It is the same with any other method; above POPULAR is used.

Does recommenderlab only deal with denser matrices? Does anyone have any ideas on how to deal with matrices with 0-1 votes, and no NAs? Thanks.

Answer 1

You need to convert your ratings matrix to a 'binaryRatingMatrix' format instead of 'realRatingMatrix' format:

ratingmatrix = as(ratingMatrix , 'binaryRatingMatrix')
recommender <- Recommender(ratingmatrix[1:1000], method="POPULAR")
predictions <- predict(recommender, ratingmatrix[1001:1011], type="topNList")