In today’s digital age, the vast amount of available content makes it increasingly challenging for users to find relevant news articles that match their interests. News recommendation algorithms play a pivotal role in alleviating this problem by providing personalized recommendations based on user preferences. Session-based recommendation systems have gained significant attention in recent years due to their ability to capture user preferences and behavior in a more granular and dynamic manner. Unlike conventional recommendation systems that rely on historical data or long-term preferences, session-based recommendation systems prioritize the user’s most recent interactions within a session. This enables instant identification of user preferences and facilitates the delivery of relevant recommendations. In this research paper, the authors explore the challenges faced by news recommendation algorithms, which play a crucial role in helping users discover relevant news articles in the era of information overload, and discuss the temporal structure of news content, the diversity of user tastes, the cold start problem faced by new users, and the significance of transformer models in the natural sciences. To address these challenges, the authors propose the utilization of session-based recommendation systems and investigate various approaches for constituting session-based recommendation systems, including the use of deep learning models, collaborative filtering, and hybrid techniques. Additionally, the authors highlighted the significance of the XLNet architecture combined with causal language modeling for accurate predictions within the session-based recommender system implementation.