Item-based recommender system with R - r

Question

Actually I'm trying to build a simple item-based recommender system and I decided to use R due to my lack of programming knowledge.
Still, some issues remain so I'll try to be as methodic as possible to explain them.
Log file
I start with a log file imported as a data frame containing many columns among which : The ID of the customers, the ID of the items and the transaction date.
Here's an overview of the log file
'data.frame': 539673 obs. of 3 variables:
$ cid: int 1 1 1 1 2 2 3 4 ...
$ iid : int 1002 1345 1346 2435 6421 4356 1002 4212 ...
$ timestamp : int 1330232453 1330233859 13302349437 1330254065 1330436051
I succeeded to turn this log file into a matrix with the customers in lines , the products in columns and the timestamp (much easier to manipulate) of the transaction date when a transaction occurs between these first two elements.
So I end up with a matrix of 100000 rows and 3000 columns, which is pretty huge.
Similarity matrix
From that point, I can create my item-based recommender system.
First, I binarize my matrix m in order to be able to compute the similarity:
mbin <- (m > 0) + 0
To compute the similarity, I use the cosine measure by a function:
getCosine <- function(x,y)
{
cosine <- sum(x*y) / (sqrt(sum(x*x)) * sqrt(sum(y*y)))
return(cosine)
}
After creating a matrix for receiving the different similarity measures,
I created two loops to measure this one:
for(i in 1:ncol(mbin)) {
for(j in 1:ncol(mbin)) {
mbin.sim[i,j] <- getCosine((mbin[i]),(mbin[j]))
}}
This similarity matrix takes too long to be computed, that is why I only focus on retrieving one particular similiarity.
Note that I've taken a random number n, I would also like to enter an item's name
n = 5
for(i in n) {
for(j in 1:ncol(mbin)) {
mbin.sim[i,j] <- getCosine(mbin[i],mbin[j])
}}
How can I achieve that?
Building and applying the recommender
From this part, I'm stuck because I can't see how I can make an easy recommender which will take one item and recommender k users to it.
Testing
Moreover, to test the recommender, I should be able to go back in time and to see if , from a certain date, I can predict the good users.
To do that, I know that I have to create a function which can give me the date of the nth transaction. Mathematically, it means that for a particular column, I have to get the nth non zero elements for that column. So I tried this but it only gives me the first one :
firstel <- function(x,n){
m <- head(x[x!=0],1)
m[n]
}
How can I do that? and moreover, How can I use this variable to discriminate between past and future events, with another function?
Sorry for this long post but I really wanted to show that I'm really into it and that I really want to overcome that step in order to begin the real analysis afterwards.
NB : I'm really doing that without complex packages due to the huge amount of data.