Datasets for recommender systems: A survey and comparative overview

Abstract Recommender systems are widely used across digital platforms such as e-commerce, media streaming, and social networks to deliver personalized content to users. The effectiveness of these systems strongly depends on the availability of high-quality datasets that capture user interactions and enable the development and evaluation of recommendation algorithms. This survey presents a comprehensive overview of publicly available datasets that are commonly used in recommender systems research. We first introduce a taxonomy of datasets based on application domain, feedback type, dataset size, and temporal characteristics. We then examine widely used datasets across several domains, including media, e-commerce, social networks, and news recommendation. In addition, a comparative analysis is conducted to highlight important dataset properties such as scale, sparsity, domain diversity, and interaction types. The paper also discusses datasets released through the RecSys Challenge, which provide large-scale industrial data for benchmarking recommendation methods. Finally, key challenges, including dataset bias, limited contextual information, and restricted data accessibility, are discussed, along with future directions for dataset development. By providing a structured overview of existing datasets and their characteristics, this survey aims to support researchers and practitioners in selecting appropriate datasets and in designing more robust and realistic recommender-systems evaluations.