The accuracy of product rating predictions is primarily determined by two factors: the product and the user. Understanding the individual and combined effects of these factors is crucial for the development of precise rating prediction systems. However, this task is complicated due to the intricate temporal dynamics associated with each user and product, as well as the cold-start problem. In this paper, we present a unique and practical method, “Time-Aware Neighborhoods and Tree Representations,” that successfully tackles these challenges. Differing from traditional methods that group users or products based on similarity, our approach groups them based on their appearance within a specific time window. We capture the temporal dynamics of users and products within each neighborhood through a structured tree representation. This not only lessens computational demands and eliminates the need to determine an optimal neighborhood size but also naturally accommodates the temporal dynamics of users and products and addresses the cold-start problem. We have developed an efficient data processing pipeline that allows the creation of neighborhoods and the building of tree structures at scale without altering the original data. Our method effectively reveals the individual and collective influences of users and products, encapsulates their temporal dynamics, and tackles the cold-start issue in a real-world, large-scale industrial data setting.