HGNN-ICFA: A Deep Learning-Based News Recommendation System Using Hybrid Graph Neural Networks and Improved Collaborative Filtering

Abstract

People rely increasingly on the internet to obtain news from a variety of sources due to the quick expansion of online information. However, this abundance leads to information overload, making it difficult for users to identify content that matches their interests. News Recommender Systems (NRS) aims to mitigate this issue by delivering personalized suggestions. An online data-mining, Deep Learning (DL)-based NRS is presented as a solution to the performance issues caused by ignoring user preferences in the recommendation process. This research proposes a hybrid architecture combining an Improved Collaborative Filtering Algorithm (ICFA) with a Hybrid Graph Neural Network (HGNN). The system module's core components are network function, database, user administration, and news recommendation. Experimental research was conducted using the News Click Behavior and Engagement Dataset from Kaggle, which contains user interaction logs including clicks, impressions, and engagement patterns across users and news articles. The data was preprocessed using normalization to scale features uniformly and enhance training stability. Additionally, Linear Discriminant Analysis (LDA) was employed for feature extraction to identify hidden topics within the news articles. The efficiency of the proposed model was evaluated based on ICFA-driven news recommendations tailored for both new and old users. The experimental findings show that the suggested method considerably enhances the metrics compared to the traditional methods on a practical dataset. The proposed model achieves 85.29% precision, 77.25% recall, and 81.87% F1 score, outperforming the robust baseline. These results confirm that the proposed HGNN-ICFA model delivers robust and personalized news recommendations across diverse user segments.