Final Project - Proposal
Dhairav Chhatbar
6/24/2020
library(recommenderlab)
library(recommenderlabBX)
library(ggplot2)
library(dplyr)
library(tidyr)
library(data.table)
library(kableExtra)General Goals
For the final project for this course I would like to use the material learned in the course to build a recommendation model on the Book-Crossing Dataset. In determining the most appropriate model to use we will:
- Simulate various recommender algorithms available in the RecommenderLab package such as Item Based Content Filtering (IBCF), User Based Content Filtering (UBCF), Singular Vector Decomposition (SVD), Principal Component Analysis (PCA), and Alternating Least Squares (ALS)
- For each algorithm, fine tune/optimize each’s parameters such as distance method, number of recommendations, and normalization method
- Evaluate each model based on ROC Curve and AUC
Once the most appropriate model is selected, I will run the model on my own profile to judge how the ratings are for me personally. I will do this by rating a number of books based on my preferences and interests and apply the recommendation model on it.
About the Dataset
The Book-Crossing dataset contains 278,858 users providing 1,149,778 ratings (explicit / implicit) about 271,379 books. This makes the dataset very sparse with most ratings at 0, and most users only have rated just 1 book.
data("BXBooks")
data("BX")
url <- "https://github.com/dhairavc/DATA612-RecommenderSystems/raw/master/Final%20Project/BX-CSV-Dump.zip"
temp <- tempfile()
download.file(url, temp)
books <- fread(unzip(temp, files = "BX-Books.csv"))
unlink(temp)
BX## 105283 x 340547 rating matrix of class 'realRatingMatrix' with 1149778 ratings.#Ratings Distribution
ratings <- getRatings(BX)
ggplot() + geom_histogram(aes(ratings), binwidth = 1, col="white", fill="pink") +
labs(title = "Ratings Distribution", x="Rating", y="Count") +
scale_y_continuous(labels = scales::comma)
#Distribution of books rated per user
ratings_by_user <- rowCounts(BX)
ratings_by_user <- data.frame(User = names(ratings_by_user), BooksRated = ratings_by_user)
ratings_by_user_sum <- ratings_by_user %>% group_by(BooksRated) %>% summarise(Users = n())
ggplot(ratings_by_user_sum, aes(y=BooksRated, x=Users)) + geom_point(col="purple")
#User with most rated books
tail(ratings_by_user_sum) %>% kable() %>% kable_styling(bootstrap_options = c("striped", "hover", "condensed"), full_width = FALSE)| BooksRated | Users |
|---|---|
| 4785 | 1 |
| 5850 | 1 |
| 5891 | 1 |
| 6109 | 1 |
| 7550 | 1 |
| 13601 | 1 |
#users with least rated books
head(ratings_by_user_sum) %>% kable() %>% kable_styling(bootstrap_options = c("striped", "hover", "condensed"), full_width = FALSE)| BooksRated | Users |
|---|---|
| 1 | 59166 |
| 2 | 12503 |
| 3 | 6533 |
| 4 | 4265 |
| 5 | 3099 |
| 6 | 2334 |
#Most read books
rating_per_book <- colCounts(BX)
rating_per_book <- data.frame(ISBN = names(rating_per_book), Rated = rating_per_book)
most_rated <- (rating_per_book %>% arrange(desc(Rated)))[1:21,]
most_rated <- left_join(most_rated, books, by = "ISBN")
knitr::include_graphics(drop_na(most_rated)$`Image-URL-M`)
