A Novel Method for News Recommendation on Websites Using the Clustered-Vectors Optimization Algorithm

This project presents a novel approach for news recommendation on websites by leveraging a clustered-vectors optimization algorithm to enhance personalization and scalability. The proposed system integrates two key models: a Content-Based Deep Recommender (CB-DR) that utilizes TF-IDF or BERT embeddings of news articles to recommend content similar to what users have previously engaged with, effectively addressing the cold-start problem by relying directly on article semantics; and a Clustering-Based Hybrid Recommender (Cluster-HybridRec) that groups users and articles based on interaction vectors, applying collaborative filtering within these clusters to reduce noise and improve recommendation diversity. The combined approach balances content-based and collaborative filtering methods to provide accurate, scalable, and diverse news recommendations. Experiments are conducted on the MIND dataset, a large-scale English news recommendation benchmark, demonstrating that the proposed clustered-vectors optimization algorithm significantly enhances recommendation relevance and user satisfaction compared to traditional methods.

Keywords: News Recommendation, Content-Based Recommender, Collaborative Filtering, Clustering, Hybrid Recommender System, TF-IDF, BERT Embeddings, Clustered-Vectors Optimization, Cold-Start Problem, MIND Dataset, Personalization, Scalability, Recommendation Diversity.