This project implements and compares two book recommendation approaches—Content-Based Filtering and Collaborative Filtering (SVD)—using the Book-Crossing dataset.

• Goal:
  Design and evaluate a recommender system pipeline using real-world book data.
• Dataset:
  Book-Crossing Dataset (Kaggle)
• Team Members:
  Shadi Farzankia 107209
  Shruti Pashine 106369
  Dharampal Singh 106316

1. Data Loading & Preprocessing:

   • Load books, ratings, and users data.
   • Clean and merge datasets, handle missing values and outliers.

2. Exploratory Data Analysis (EDA):

   • Visualize distributions, check for anomalies, and understand feature relationships.

3. Recommendation Approaches:

   • Content-Based Filtering: Uses book metadata (title, author, publisher) with TF-IDF and cosine similarity.
   • Collaborative Filtering (SVD): Uses user-book ratings and matrix factorization (Surprise SVD).

4. Evaluation:

   • Precision@5 (Hit Rate) for both methods.
   • RMSE for SVD.

5. Comparison & Discussion:

   • Compare strengths, weaknesses, and visualize results.

Place the following files in a data/ directory:

• books.csv
• ratings.csv
• users.csv

1. Clone this repository.
2. Install dependencies:

   pip install -r requirements.txt

3. Open the notebook (Code/RecommenderSytems.ipynb) in Jupyter or VS Code.
4. Run all cells in order.

• Content-Based Filtering:
  • Interpretable, works for new/unpopular books, higher hit rate.
• SVD Collaborative Filtering:
  • More accurate in rating prediction (lower RMSE), more personalized, but needs enough user-book interactions.

• Data sparsity and cold-start issues for collaborative filtering.
• Evaluation for SVD is limited to a sample of users for computational reasons.
• Future work: hybrid models, more features, deep learning approaches.

@Shruti Pashine, @Shadi Farzankia, @Dharampal Singh

Dataset: Book-Crossing Dataset (Kaggle)