Task 1: Read Prediction

Baseline Approach

Methodology:
Followed the approach outlined in baselines.py, which predicts whether a user would read a book based on its popularity.

Steps:
Calculated a popularity threshold using the top 50% of book interactions in the training data.

For test pairs:
Predicted 1 (indicating the user would read the book) if the book's popularity exceeded the threshold.
Predicted 0 otherwise.

Improved Model
Objective: Enhance prediction accuracy by incorporating additional user and book features.

Methodology:
Trained a logistic regression model with the following features:
User Read Count: Total number of books the user has read.
Book Popularity: Total number of times each book has been read.
Logistic regression was chosen as a simple and effective classification model for binary prediction tasks.
The model was trained on features extracted from the training data and used to predict read actions for test pairs in pairs_Read.csv.

Task 2: Rating Prediction

Baseline Approach

Methodology:
Used the global average rating as the default prediction.
Enhanced this by incorporating user-specific averages:
If the user had historical ratings in the training data, their average rating was used instead of the global average.
This approach accounts for user-level biases and provides a straightforward baseline.

Improved Model
Objective: Improve prediction accuracy by modeling latent user-book interactions.
Methodology:
Implemented Matrix Factorization with Singular Value Decomposition (SVD) to decompose the user-book interaction matrix into latent factors.
Steps:
Set the number of latent factors (k) to 20 to balance model complexity and performance.
Latent factors represent hidden characteristics of users and books, enabling personalized predictions.
Incorporated user and book biases implicitly through the latent factor decomposition.
For test pairs in pairs_Rating.csv:
Predicted ratings using the decomposed matrix when user and book data were available.
Defaulted to global or user-specific average ratings as fallback for unseen users or books.

Summary

Baseline Models
Provided a solid starting point for understanding the dataset and tasks.
Focused on simple heuristics like popularity thresholds for read prediction and global/user averages for rating prediction.

Enhanced Models
Read Prediction:
Used logistic regression with user and book features, providing a more nuanced approach than the popularity threshold.
Rating Prediction:
Leveraged matrix factorization (SVD) for collaborative filtering, capturing user preferences and book characteristics.
Improved generalization to unseen user-book pairs by introducing latent factors.

These improvements address key limitations in the baselines, balancing interpretability with predictive accuracy. The combination of feature engineering and collaborative filtering provides robust predictions for both tasks.