Book Recommendation System

A book recommendation system conducted in a machine learning class at Gachon University.

INDEX

• Business Objective
• Data Exploration
• Data Preprocessing
• Modeling
• Evaluation / Analysis

Business Objective

• Enhancing User Experience : analyze each user’s reading preferences
• Providing Accurate and Efficient Recommendations : By employing SVD, user-based, and item-based collaborative filtering models --> Personalized reading experience

Data Exploration

Goodbooks-10k Dataset:

• books_tag.csv: goodreads_book_id, tag_id, count
• books.csv: id, book_id, best_book_id, work_id, books_count, isbn13, authors, original_publication_year, original_title
• ratings.csv: book_id, user_id, rating
• tags.csv: tag_id, tag_name
• to_read.csv: user_id, book_id

Book Recommendation Dataset:

• Books.csv: ISBN, Book_Title, Book-Author, Year_Of-Publication, Publisher, Image-URL-S, Image-URL-M, Image-URL-L
• Ratings.csv: User-ID, ISBN, Book-Rating
• Users.csv: User-ID, Location, Age

Data Preprocessing

Goodbooks-10k

• Remove unnecessary columns
• Merge ratings.csv, books.csv based on book_id
• Fill the missing values

Book Recommendation

• Remove unncessary columns
• Convert all string data to lowercase
• Merge Ratings.csv, Books.csv and Users.csv based on ISBN
• Delete noise data by eliminating data whose rating is 0.

Filtering

• Select books with 10 or more ratings
• Select books that have been rated by at least 5 different users.

Normalization

• Unify the column name of both data sets
• Change the data type of ISBN and User_ID
• Map the ISBN based on book title
• Fix the size of ISBN
• Exclude users only rated one book
• Exclude duplicated books of both dataset

Modeling

• Implemented using Surprise library

SVD

• SVD is a dimensionality reduction technique that analyzes latent factors between users and items to generate recommendations
• Recommendations are generated based on the provided user ID
• Output: displays the title of each recommended book and its predicted rating

User-based

• User-Based Collaborative Filtering finds similar users based on Pearson correlation
• Recommends books that similar users have rated highly
• Output: provides a list of books with predicted ratings for the specified user

Item-based

• Item-Based Collaborative Filtering compares items to find similarities based on user ratings
• Books similar to those the user has already rated highly are recommended
• Output: generates a list of books with predicted ratings tailored to the specified user

Evaluation / Analysis

Team Member

• Kwon Soyeong : data modeling
• Song Eunjung : data modeling
• Jeon Sihyeon : data evaluation
• Jang Yonggeun : data preprocessing
• Kim Jeongsu : data preprocessing