Simulated User Behavior for Recommender Systems Applied to the MIND Dataset

The SUBER (Simulated User Behavior for Recommender Systems) framework, a novel approach in the field, has been extended and applied to the MIND (Microsoft News Dataset) to improve the performance of news recommender systems. This study addresses the need for more personalized news recommendations by leveraging SUBER to simulate realistic user interactions, especially in areas with sparse data. Utilizing the Stable Baselines3 library, extensive trials with various models were conducted to identify the most effective configurations. Our approach involved generating synthetic interaction data through simulations, which were then used to train and evaluate different recommendation models. Although the primary goal was to get SUBER to work effectively with the MIND dataset, the extension successfully integrated and operated within this new context. This study underscores the potential of advanced simulation techniques, like those provided by SUBER, to enhance the capabilities of news recommendation systems. Key achievements include successfully applying SUBER to the MIND dataset and demonstrating its robustness and adaptability in a new domain. This work sets a new benchmark for future research in news recommendation systems. It contributes to the broader field of recommender systems by showcasing the utility of user behavior simulation in optimizing recommendation algorithms, highlighting the importance of using simulated data to complement fundamental user interactions for improved user experiences and satisfaction with recommended content.