Measuring the Predictability of Recommender Systems using Structural Complexity Metrics

As the amount of information and content available to users continues to explode, recommender systems play an essential role in enhancing users' experience in areas ranging from e-commerce and entertainment to social media and personalized content delivery. These systems are designed to balance the huge amount of content available with the individual preferences of users to maximize the interaction-utility ratio of the users. Among the various paradigms in recommendation systems, collaborative filtering (CF) stands out as a widely adopted approach known for its effectiveness in delivering valuable and personalized recommendations to users Shi et al. (2014). By leveraging the collective wisdom of users' preferences and behaviors, collaborative filtering recommends items based on the similarity of users' tastes and interactions. Despite its practical success, much of the knowledge surrounding collaborative filtering remains largely empirical, leaving a gap in our comprehensive understanding of the underlying characteristics of the filtering problem and the intricacies of this specific approach. Unraveling the inner workings of collaborative filtering is a major challenge due to its inherent complexity. The interactions between users and items within a recommendation system generate large and intricate datasets, making extracting meaningful patterns and underlying mechanisms difficult. To address these challenges, researchers are increasingly turning to interdisciplinary approaches that combine insights from data science, machine learning, and the social sciences Chen et al. (2023). By integrating theories and methods from these diverse fields, they aim to gain a more holistic understanding of how users' social interactions, psychology, and preferences influence the collaborative filtering process.