Book Recommendations

The goal of this project is to build a recommender system that suggests relevant books based on the person's interests.

The Streamlit app takes a user ID as input and displays a list of 10 recommended books.

Dataset

I used the goodbooks-10k dataset created by Zygmunt Zając.

Data includes:

• 10,000 books (different editions of the book have the same book ID)
• 53,424 users
• 6M ratings (ratings are whole numbers from 1 to 5)

Approach

1. Data Exploration
2. Modeling - collaborative filtering using Surprise library
   • Compare different Surprise algorithms on the training set (cross-validation)
   • Evaluate shortlisted models on the test set and choose the final model
3. Create a Streamlit app that recommends 10 books for the user

Models

1. Normal Predictor (random predictions based on rating distribution) - as a baseline model

   • Test MAE = 1.05

2. k-NN Baseline, item-based - the best model

   • Test MAE = 0.60
   • The Mean Absolute Error on the test set has been reduced by 42% compared to the Normal Predictor (from 1.05 to 0.60)
   • On average, the rating predictions by k-NN model are off by 15% (mean rating = 3.92)

The k-NN Baseline model is used in the app (app_book_recommender_1.py) to predict unknown ratings for the user. Then the books with the highest predicted ratings are included in the recommendation list.