A multi-objective optimization approach for session-based recommendation systems

Recommender systems face the persistent challenge of balancing multiple conflicting objectives, such as relevance, diversity, and user engagement, while adapting to the complexities of session-based data. Traditional methods often struggle to address these challenges effectively, particularly when user interactions are diverse, sparse, or structured in short sessions. Moreover, the trade-offs between accuracy-focused metrics and diversity-oriented metrics pose additional hurdles in achieving well-rounded recommendations. This paper presents a novel session-based multi-objective recommendation approach designed to address these challenges. The method employs session clustering to group similar behavioral patterns, reducing complexity and enhancing focus. Within each cluster, focused item subsetting refines the recommendation space, enabling efficient identification of high-performing solutions. Additionally, Differential Evolution (DE)-based optimization facilitates a balance between competing objectives, while cross-cluster knowledge sharing accelerates convergence and enhances generalization by transferring effective strategies between session clusters. Extensive experiments conducted on real-world datasets demonstrate that the proposed approach consistently outperforms state-of-the-art session-based and multi-objective optimization-based recommender systems. These results highlight its ability to adapt to diverse interaction patterns and effectively balance conflicting objectives, making it a practical and scalable solution for modern recommendation systems.