This project implements a hybrid book recommendation system combining Content-Based Filtering and a Model-Based Collaborative Filtering approach using Singular Value Decomposition (SVD). This system provides book recommendations by leveraging both item content features and latent user/item taste profiles learned from rating patterns.

The solution is implemented in Python and utilizes the pandas, scikit-learn, and scikit-surprise libraries.

1. Clone the Project Repository: First, clone this project's repository to your local machine:

   git clone https://github.com/aayushrautela/Book-Recommendation-System.git
   cd Book-Recommendation-System

2. Prerequisites (Python Libraries):

   • Python (version 3.10 or 3.11 recommended for library compatibility).
   • The required Python libraries are listed in requirements.txt.
   • It is recommended to use a virtual environment.
   • Install the dependencies using pip:

   pip install -r requirements.txt

3. Dataset (goodbooks-10k):

   • Download: This project uses the goodbooks-10k dataset. You need to download it from its original GitHub repository: https://github.com/zygmuntz/goodbooks-10k
   • Placement:
     • After downloading, create a subdirectory named data/ in the root of this cloned project.
     • Extract the downloaded dataset and place the following essential CSV files into this data/ subdirectory:
       • books.csv
       • ratings.csv
       • tags.csv
       • book_tags.csv

This system requires a "training" step before recommendations can be made.

The SVD model needs to be trained on the rating data first. This process will learn the latent factors and save a cf_svd_model.joblib file (a "pretrained checkpoint").

• To train the model with the optimized default parameters, run:

  python collaborative_recommender.py train

• This will run for a set number of epochs and only needs to be done once, or whenever you want to retrain the model.

Once the model is trained, you can get recommendations using the following scripts.

1. Hybrid Recommender (hybrid_recommender.py) - Recommended

   • This is the main script. It loads the pretrained CF model and the CB model to provide blended recommendations for a specific user.
   • To run, provide a user_id:

     python hybrid_recommender.py <user_id>

     Example: python hybrid_recommender.py 100

2. Standalone Content-Based Recommender (content_based_recommender.py)

   • This script provides recommendations based only on content similarity to a specific book.
   • To run, provide a book_id:

     python content_based_recommender.py <book_id>

     Example: python content_based_recommender.py 6

3. Standalone Collaborative Filtering Recommender (collaborative_recommender.py)

   • This script uses the pretrained SVD model to provide recommendations based only on collaborative filtering.
   • To run, use the recommend action and provide a user_id:

     python collaborative_recommender.py recommend --user_id <user_id>

     Example: python collaborative_recommender.py recommend --user_id 100

To find the best parameters for the SVD model, you can run the evaluation script.

• This script uses GridSearchCV to test multiple parameter combinations and will print the best ones based on RMSE. It saves the results to tuning_results.csv and generates a heatmap plot tuning_heatmap.png.
• Warning: This process is very time-consuming.
• To run:

  python evaluate.py

• content_based_recommender.py: Contains all logic for the content-based filtering approach.
• collaborative_recommender.py: Implements the SVD model-based collaborative filtering, including training and inference.
• hybrid_recommender.py: Combines the two models to produce final recommendations.
• evaluate.py: Used for hyperparameter tuning and evaluation of the SVD model.
• data/: Directory where the dataset CSV files must be placed.
• requirements.txt: Lists all Python dependencies.