Hybrid Artificial Intelligence Approaches for Cold-Start Mitigation in Personalized News Recommendation: A Survey
TL;DR
The findings indicate that Hybrid AI has become the dominant research direction because it improves recommendation accuracy, semantic understanding, personalization, and robustness under sparse interaction conditions.
Abstract
The rapid expansion of digital news platforms has intensified information overload, making personalized news recommendation systems essential for delivering relevant content to users. However, their effectiveness is constrained by the cold-start problem arising from insufficient interaction data for new users, new news articles, or newly deployed platforms. This paper presents a survey of Hybrid Artificial Intelligence (AI) approaches for mitigating the cold-start problem in personalized news recommendation. Recent studies published between 2020 and 2025 were thematically reviewed to examine the evolution of recommendation techniques from traditional Collaborative Filtering and Content-Based Filtering to Hybrid AI frameworks incorporating Deep Learning, Graph Neural Networks, Knowledge Graphs, Transformer models, Large Language Models, and zero-shot learning. The findings indicate that Hybrid AI has become the dominant research direction because it improves recommendation accuracy, semantic understanding, personalization, and robustness under sparse interaction conditions. User cold-start remains the most widely investigated challenge, while the MIND dataset and metrics such as Precision, Recall, AUC, nDCG, and MRR are the most frequently adopted evaluation standards. The survey highlights current research trends, identifies unresolved challenges, and provides directions for developing more scalable, explainable, and efficient Hybrid AI-based personalized news recommendation systems.