NexLetter : Personalized News Recommendation System

Abstract

This thesis presents the design and implementation of a hybrid newsletter recommendation system that personalizes article suggestions based on user preferences, interaction behavior, and semantic similarity between article titles. The system retrieves news articles from an external API and uses a scoring function to rank them by relevance, combining country and category matches, liked articles, and title similarity via TF-IDF and cosine similarity. The backend is developed using Python, PostgreSQL, and FastAPI, and follows a modular architecture with scheduled article ingestion and persistent interaction logging. The system was evaluated through real-user testing and outperformed a random baseline in both click-through rate (CTR) and precision at 5 (P@5), showing its effectiveness in improving content relevance. The project lays the groundwork for future enhanc