Book Rating Prediction Model

In this project, we implemented the use of a machine-learning model in order to predict book ratings. The steps were as follows:

• Importing & Cleaning the Dataset
• Exploratory Data Analysis (EDA)
• Feature Engineering
• Modelling

Dataset

The original dataset for this project was obtained from Goodreads book metadata. It is titled 1.books_original.csv. The columns are as follows:

Data Columns	Description
bookID	A unique identification number for a book.
title	The name under which the book was published. It may include the book universe and volume number if it is part of a series
authors	The name of the author(s) and additional contributors for a book. Multiple authors/contributors are delimited with a '/'
average_rating	The average rating a book received in total
isbn	Another unique identification number for a book. It stands for International Standard Book Number.
isbn13	Another unique identification number for a book. It is a 13-digit ISBN to identify a book, instead of the standard 11-digit ISBN.
language_code	A unique identifier for the language of a book e.g. 'eng' is standard for 'English'.
num_pages	The number of pages a book contains.
ratings_count	The total number of ratings a book received.
text_reviews_count	The total number of written text reviews a book received.
publisher	The name of the publisher for a book.

The original dataset was then merged to an additional dataset which was obtained from Mengting's Goodreads 2017 Data Dump. This merged dataset is found under 3.books_merged_series_format.csv.

Mengting Wan, Julian McAuley, "Item Recommendation on Monotonic Behavior Chains", in RecSys'18. [bibtex] Mengting Wan, Rishabh Misra, Ndapa Nakashole, Julian McAuley, "Fine-Grained Spoiler Detection from Large-Scale Review Corpora", in ACL'19. [bibtex]

Two columns were then added to the dataset with the use of a small web-scraping script to obtain data from Open Library's API. These are found in the 3.books_merged_series_format.csv file:

Data Columns	Description
format	The format for a book.
series exists	Indicates 1 if the book is part of a series, else indicates 0.

Exploratory Data Analysis (EDA)

We mainly explored the relation of each feature to the target column i.e. average ratings column. We also looked at outliers for each feature.

Feature Engineering

Based on how our EDA, we retained the num_pages, ratings_count, text_reviews_count and a calculated column called ratings_reviews_ratio.

Modelling

We chose a regression model type to predict the average rating. We tested out four different models:

• Linear Regression
• Polynomial Regression
• Random Forest
• AdaBoost & Decision Tree

Outcome

The Random Forest model outperformed all other models, with its predictions being relatively close to the actual values on average. Notably, the model returned a consistent R2 score of around 0.8.

Getting Started

Software Prerequisites

• Windows
• Python 3.10 Environment
• Jupyter Notebook

Software Installation

• Open Anaconda Prompt
• Type and execute conda create --name myenv python=3.10
• Type and execute requirements.txt to create an environment from the requirements.txt file.
• Open the Jupyter notebook ML_final