This recommender engine finds user movie genre preference profiles that are most closely correlated to the target user’s genre preference profile and returns suggestions based on the highest mean ratings of unwatched movies from these similar users. The 100k row version of the Movielens dataset used for this project is based on anonymous user reviews and viewing history collected in the late 1990s and is available here.
library(Hmisc)
library(dplyr)
movies <- read.csv('http://files.grouplens.org/datasets/movielens/ml-100k/u.data', sep = "\t", header=F)
colnames(movies) <- c("user_id", "movie_id", "rating", "timestamp")
items <- read.csv('http://files.grouplens.org/datasets/movielens/ml-100k/u.item', sep = "|", header=F)
names <- tolower(Cs(movie_id, title, release_date, video_release_date, imdb, unknown, Action, Adventure, Animation, Childrens, Comedy, Crime, Documentary, Drama, Fantasy, Noir, Horror, Musical, Mystery, Romance, SciFi, Thriller, War, Western))
colnames(items) <- names
df <- left_join(movies, items, by = "movie_id")
dfIndiv <- df[, c(1, 9:27)]
userTotals <- dfIndiv %>% group_by(user_id) %>% summarise_all(funs(sum))
userTrans <- t(userTotals[, 2:19])
colnames(userTrans) <- 1:943The data is collected and properly formatted. Now to build the function:
movieRecommender <- function(x) {
similarUsers <- function(x) {
user <- userTrans[, x]
others <- userTrans[, -x]
# correlate target user profile with other user profiles
comparisons <- data.frame(t(cor(user, others)))
comparisons$user <- colnames(others)
colnames(comparisons) <- c("correlation", "user")
comparisons <- comparisons %>%
group_by(user) %>%
arrange(desc(correlation))
tenMostSimilar <- comparisons[1:10, 2]
# return 10 most highly correlated user profiles
return(tenMostSimilar)
}
ten <- similarUsers(x)
ten <- as.numeric(unlist(ten))
# list all movies target user has watched
userWatched <- movies %>%
filter(user_id == x) %>%
select(movie_id, rating)
# list all movies the 10 most similar users have watched
otherUserMovies <- movies %>%
filter(user_id %in% ten) %>%
select(movie_id, rating)
# anti-join to get movies not yet seen by target user
unwatchedMovies <- anti_join(otherUserMovies, userWatched, by = "movie_id")
# find the highest mean ratings of unwatched movies
highestRated <- unwatchedMovies %>%
group_by(movie_id) %>%
summarise(averages = mean(rating)) %>%
arrange(desc(averages))
# find the most frequently watched movies in unwatchedMovies
mostFrequent <- unwatchedMovies %>%
group_by(movie_id) %>%
summarise(watched = n()) %>%
arrange(desc(watched))
ratedAndWatched <- inner_join(highestRated, mostFrequent, by = "movie_id")
# find the highest rated, most watched movies from similar users
ratedAndWatched <- left_join(ratedAndWatched, items, by = "movie_id")
userFavGenres <- userTrans[, x]
# find the top 5 favorite genres of the target user
userFavGenres <- userFavGenres %>% sort() %>% tail(5) %>% names()
genreNames <- unlist(strsplit(userFavGenres, "_tot"))
ratedAndWatched$keep <- ifelse(ratedAndWatched[, genreNames[1]] == 1 |
ratedAndWatched[, genreNames[2]] == 1 |
ratedAndWatched[, genreNames[3]] == 1 |
ratedAndWatched[, genreNames[4]] == 1 |
ratedAndWatched[, genreNames[5]] == 1, 1, 0)
ratedAndWatched <- ratedAndWatched %>% filter(keep == 1)
# return the 100 most watched highest rated films not seen by target user
highestAvgs <- ratedAndWatched %>% arrange(desc(watched))%>% head(100)
highestAvgs <- highestAvgs %>% arrange(desc(averages)) %>% head(10)
movieSuggestions <- head(highestAvgs$title, 10)
return(movieSuggestions)
}Testing the recommender engine:
# genre preference history for user #6
t(userTotals[6, ]) ## [,1]
## user_id 6
## unknown 0
## action 25
## adventure 22
## animation 10
## childrens 20
## comedy 66
## crime 14
## documentary 1
## drama 104
## fantasy 3
## noir 6
## horror 4
## musical 13
## mystery 12
## romance 41
## scifi 13
## thriller 24
## war 21
## western 5It looks like user six’s top five genres are dramas, comedies, romance, action, and thrillers. Let’s see the top ten movies the function recommends:
movieRecommender(6) # user_id 6## [1] Chinatown (1974)
## [2] Some Folks Call It a Sling Blade (1993)
## [3] It Happened One Night (1934)
## [4] Being There (1979)
## [5] Singin' in the Rain (1952)
## [6] Strictly Ballroom (1992)
## [7] Raise the Red Lantern (1991)
## [8] Rear Window (1954)
## [9] Bullets Over Broadway (1994)
## [10] Ridicule (1996)
## 1664 Levels: 'Til There Was You (1997) ...Now we’ll test the function on user 21:
# genre preference history for user #21
t(userTotals[21, ])## [,1]
## user_id 21
## unknown 0
## action 39
## adventure 16
## animation 9
## childrens 12
## comedy 32
## crime 15
## documentary 1
## drama 47
## fantasy 2
## noir 2
## horror 61
## musical 7
## mystery 9
## romance 15
## scifi 23
## thriller 66
## war 6
## western 0It looks like user twenty-one’s top five genres are thrillers, horror, drama, action, and comedy. Let’s see the top ten movies the function recommends:
movieRecommender(21) ## [1] Aliens (1986)
## [2] Braveheart (1995)
## [3] Monty Python and the Holy Grail (1974)
## [4] Raging Bull (1980)
## [5] Apollo 13 (1995)
## [6] Professional, The (1994)
## [7] Faster Pussycat! Kill! Kill! (1965)
## [8] Fugitive, The (1993)
## [9] Big Night (1996)
## [10] Monty Python's Life of Brian (1979)
## 1664 Levels: 'Til There Was You (1997) ...