Building a movie recommender in R

library(tidyr)
#users ID range from 1 to 5, movie names are A to E and ratings are 1 to 5
ratings <- data.frame(
  user = c(1, 1, 2, 2, 3, 3, 4, 4, 5, 5),
  movie = c("A", "B", "A", "C", "B", "C", "D", "E", "D", "E"),
  rating = c(5, 4, 4, 3, 3, 2, 2, 1, 1, 5)
)
# Create a user-item matrix; output dimension is 5 by 5 due to 5 users and 5 different movies
user_item_matrix <- spread(ratings, movie, rating, fill = 0)
#The rows of the matrix represent the user number and the columns representing each movie.
print(user_item_matrix)

#cosine similarity function
cosine_similarity <- function(user1, user2) {
  dot_product <- sum(user1 * user2)
  norm_user1 <- sqrt(sum(user1^2))
  norm_user2 <- sqrt(sum(user2^2))
  
  if (norm_user1 == 0 || norm_user2 == 0) {
    return(0)  # Handling division by zero error
  }
  
  return(dot_product / (norm_user1 * norm_user2))
}

# Function to recommend movies for a given user ID
recommend_movies <- function(user_id, user_item_matrix) {
  user_ratings <- user_item_matrix[user_id, ]
  
  # Calculate cosine similarity between the selected user and all other users
  similarities <- sapply(1:nrow(user_item_matrix), function(i) {
    cosine_similarity(user_ratings, user_item_matrix[i, ])
  })
  # Sorting users by similarity in descending order
  similar_users <- order(similarities, decreasing = TRUE)
  
  # Find movies the user has not seen (i.e with a rating = 0)
  unseen_movies <- which(user_ratings == 0)
  
  # Recommend movies from most similar users; using the union operator to avoid duplicate movies being returned 
  recommended_movies <- numeric(0)
  for (user in similar_users) {
    similar_user_ratings <- user_item_matrix[user, ]
    recommended_movies <- union(recommended_movies, unseen_movies[unseen_movies %in% which(similar_user_ratings > 0)])
    
    # Limit the number of recommendations (here we are limiting the number of recommendations to five)
    if (length(recommended_movies) >= 5) {
      break
    }
  }
  #returning the list of recommended movies for a given user ID
  return(names(user_ratings)[recommended_movies])
}

library(tidyr)
# Sample ratings data
ratings <- data.frame(
  user = c(1, 1, 2, 2, 3, 3, 4, 4, 5, 5),
  movie = c("A", "B", "A", "C", "B", "C", "D", "E", "D", "E"),
  rating = c(5, 4, 4, 3, 3, 2, 2, 1, 1, 5)
)
# Create a user-item matrix
user_item_matrix <- spread(ratings, movie, rating, fill = 0)
# Cosine similarity function
cosine_similarity <- function(user1, user2) {
  dot_product <- sum(user1 * user2)
  norm_user1 <- sqrt(sum(user1^2))
  norm_user2 <- sqrt(sum(user2^2))
  
  if (norm_user1 == 0 || norm_user2 == 0) {
    return(0)  # Handling division by zero
  }
  
  return(dot_product / (norm_user1 * norm_user2))
}
# Function to recommend movies for a given user id
recommend_movies <- function(user_id, user_item_matrix) {
  user_ratings <- user_item_matrix[user_id, ]
  
  # Calculate similarity between the selected user and all other users
  similarities <- sapply(1:nrow(user_item_matrix), function(i) {
    cosine_similarity(user_ratings, user_item_matrix[i, ])
  })
  
  # Exclude the user itself
  similarities[user_id] <- -Inf
  
  # Sort users by similarity in descending order
  similar_users <- order(similarities, decreasing = TRUE)
  
  # Find movies the user has not seen (rating = 0)
  unseen_movies <- which(user_ratings == 0)
  
  # Calculate predicted ratings for unseen movies
  movie_scores <- numeric(ncol(user_item_matrix) - 1)  # Subtract 1 to ignore user column
  names(movie_scores) <- colnames(user_item_matrix)[-1]
  
  for (movie_index in unseen_movies) {
    movie <- colnames(user_item_matrix)[movie_index]
    weighted_sum <- 0
    similarity_sum <- 0
    
    for (user in similar_users) {
      if (user_item_matrix[user, movie_index] > 0) {
        weighted_sum <- weighted_sum + similarities[user_id] * user_item_matrix[user, movie_index]
        similarity_sum <- similarity_sum + abs(similarities[user_id])
      }
    }
    
    if (similarity_sum > 0) {
      movie_scores[movie] <- weighted_sum / similarity_sum
    }
  }
  
  # Recommend movies with the highest predicted ratings
  recommended_movies <- names(sort(movie_scores, decreasing = TRUE))
  
  return(recommended_movies)
}
# Test the function for User 1
user_id <- 1
recommended_movies <- recommend_movies(user_id, user_item_matrix)
print(paste("Recommended movies for User", user_id, ":"))
print(recommended_movies)