This recommender system will use MovieLense data from recommenderlab package. This system will recommend top 5 movies to randomly selected users using item-based collaborative filtering.
## Loading required package: Matrix## Loading required package: arules##
## Attaching package: 'arules'## The following objects are masked from 'package:base':
##
## abbreviate, write## Loading required package: proxy##
## Attaching package: 'proxy'## The following object is masked from 'package:Matrix':
##
## as.matrix## The following objects are masked from 'package:stats':
##
## as.dist, dist## The following object is masked from 'package:base':
##
## as.matrix## Loading required package: registry##
## Attaching package: 'dplyr'## The following objects are masked from 'package:arules':
##
## intersect, recode, setdiff, setequal, union## The following objects are masked from 'package:stats':
##
## filter, lag## The following objects are masked from 'package:base':
##
## intersect, setdiff, setequal, union## Loading required package: SnowballC##
## Attaching package: 'lsa'## The following object is masked from 'package:dplyr':
##
## querydata("MovieLense")
movies <- as.matrix(MovieLense@data)
dim(movies)## [1] 943 1664This data contains 943 users with ratings on 1664 movies.
The dataset will be normalized taking standard score using universal mean.
movies[movies==0] <- NA
mmean <- mean(movies, na.rm = T)
msd <- sd(movies, na.rm=T)
moviesnorm <- (movies - mmean)/msdWe will be generating similarity matrix based on similarities between items.
library(lsa)
movietemp <- movies
movietemp[is.na(movietemp)] = 0
movie_sim_mtx = cosine(as.matrix(movietemp))
#image(as.matrix(movie_sim_mtx))
#library(corrplot)
#corrplot(as.matrix(movie_sim_mtx), is.corr = FALSE, method = "color")Populate all na values for a user
getTopFive <- function(u){
# get all normalized ratings for this user
user_ratings <- moviesnorm[u,]
# get list of movies user did not rate
movies_user_not_rated <- which(is.na(user_ratings))
# get list of movies user rated
movies_user_rated <- which(!is.na(user_ratings))
# list of movie names
movie_names <- names(user_ratings)
# list to store predicted values for not rated moves
now_rated_value <- list()
now_rated_name <- list()
# calculated weighted similarity for non related
for(i in 1:length(movies_user_not_rated)){
# get column from similarity matrix
movie_item_simlarity = movie_sim_mtx[i,]
# calculate weight and weight average
weight_total = sum(movie_item_simlarity[unlist(movies_user_rated)])
weighted <- movie_item_simlarity * moviesnorm[1,]
# store the movie rating and its name
now_rated_value[i] <- sum(weighted,na.rm=T)/weight_total
now_rated_name[i] <- movie_names[movies_user_not_rated[i]]
}
# return top five recommended values
now_rated_df <- cbind.data.frame("m"=unlist(now_rated_name),
"v"=unlist(now_rated_value),
stringsAsFactors = FALSE)
now_rated_sorted_df <- now_rated_df[order(-now_rated_df$v),]
rv <- head(now_rated_sorted_df$m,n=5)
return(rv)
}# Print out top 5 recommendations for 10 users
recommends <- data.frame(matrix(nrow=10,ncol=6))
names(recommends) <- c("User","Top1","Top3","Top3","Top4","Top5")
# randomly pick 10 users
set.seed(123)
users <- sample(1:nrow(moviesnorm),10, replace = FALSE)
for(i in c(1:length(users))){
top5 <- getTopFive(users[i])
recommends[i,] <- c(users[i],as.list(top5))
}
# print out the recommendations
kable(recommends)| User | Top1 | Top3 | Top3 | Top4 | Top5 |
|---|---|---|---|---|---|
| 272 | Good Morning (1971) | Locusts, The (1997) | Anna (1996) | Jupiter’s Wife (1994) | Return of the Pink Panther, The (1974) |
| 743 | Buddy (1997) | Big One, The (1997) | Favor, The (1994) | Prefontaine (1997) | Dadetown (1995) |
| 385 | Smilla’s Sense of Snow (1997) | Silence of the Palace, The (Saimt el Qusur) (1994) | Underground (1995) | Price Above Rubies, A (1998) | Jean de Florette (1986) |
| 831 | Jury Duty (1995) | American Dream (1990) | Mirage (1995) | Gone with the Wind (1939) | Two Friends (1986) |
| 884 | Family Thing, A (1996) | 8 Heads in a Duffel Bag (1997) | Collectionneuse, La (1967) | New Age, The (1994) | Hana-bi (1997) |
| 43 | Rhyme & Reason (1997) | Hotel de Love (1996) | Vermont Is For Lovers (1992) | Bitter Sugar (Azucar Amargo) (1996) | It Takes Two (1995) |
| 495 | Enfer, L’ (1994) | JLG/JLG - autoportrait de decembre (1994) | Boys, Les (1997) | Clean Slate (Coup de Torchon) (1981) | Secret Adventures of Tom Thumb, The (1993) |
| 836 | Stranger, The (1994) | Boys, Les (1997) | Condition Red (1995) | Little City (1998) | Willy Wonka and the Chocolate Factory (1971) |
| 516 | 1-900 (1994) | Dunston Checks In (1996) | Story of Xinghua, The (1993) | Bitter Sugar (Azucar Amargo) (1996) | Nowhere (1997) |
| 427 | Kazaam (1996) | Larger Than Life (1996) | Angel Baby (1995) | Mostro, Il (1994) | Hana-bi (1997) |
# collect movie names and its index
movie_names <- colnames(movies)
# run recommenderlab builtin function
recsys <- Recommender(data = MovieLense, method = "IBCF", parameter = list(method = "Cosine"))
recsysTop5 <- predict(object = recsys, newdata = MovieLense, n = 5)
top5items <- recsysTop5@items
# get top 5 movie names
top5builtin <- movie_names[unlist(top5items[22])]
# Print out top 5 recommendations for 10 users
recommends_builtin <- data.frame(matrix(nrow=10,ncol=6))
names(recommends_builtin) <- c("User","Top1","Top3","Top3","Top4","Top5")
for(i in c(1:length(users))){
top5 <- movie_names[unlist(top5items[users[i]])]
recommends_builtin[i,] <- c(users[i],as.list(top5))
}
kable(recommends_builtin)| User | Top1 | Top3 | Top3 | Top4 | Top5 |
|---|---|---|---|---|---|
| 272 | Postino, Il (1994) | From Dusk Till Dawn (1996) | Crimson Tide (1995) | Eat Drink Man Woman (1994) | Three Colors: Blue (1993) |
| 743 | Usual Suspects, The (1995) | Muppet Treasure Island (1996) | Eat Drink Man Woman (1994) | Hoop Dreams (1994) | Pulp Fiction (1994) |
| 385 | Return of the Pink Panther, The (1974) | FairyTale: A True Story (1997) | Midnight in the Garden of Good and Evil (1997) | Hard Rain (1998) | Wonderland (1997) |
| 831 | From Dusk Till Dawn (1996) | Rumble in the Bronx (1995) | Batman Forever (1995) | To Wong Foo, Thanks for Everything! Julie Newmar (1995) | Natural Born Killers (1994) |
| 884 | GoldenEye (1995) | Richard III (1995) | Mighty Aphrodite (1995) | From Dusk Till Dawn (1996) | Muppet Treasure Island (1996) |
| 43 | Mighty Aphrodite (1995) | Crumb (1994) | Haunted World of Edward D. Wood Jr., The (1995) | Supercop (1992) | Sleeper (1973) |
| 495 | French Twist (Gazon maudit) (1995) | Crumb (1994) | Desperado (1995) | Faster Pussycat! Kill! Kill! (1965) | Carlito’s Way (1993) |
| 836 | GoldenEye (1995) | Four Rooms (1995) | Dead Man Walking (1995) | From Dusk Till Dawn (1996) | Muppet Treasure Island (1996) |
| 516 | Four Rooms (1995) | Twelve Monkeys (1995) | Postino, Il (1994) | Batman Forever (1995) | Eat Drink Man Woman (1994) |
| 427 | Twelve Monkeys (1995) | Dead Man Walking (1995) | Mr. Holland’s Opus (1995) | Faster Pussycat! Kill! Kill! (1965) | Brother Minister: The Assassination of Malcolm X (1994) |
From running my own recommender algorithm and built-in, the built-in processes the feed much faster. Unfortunately, the outputs from both systems do not match which tells me that should improve on the system I have created.