Conceptual Foundations of Cross-Domain Recommender Systems: An Ontological Perspective

Abstract

This paper presents a comprehensive synthesis of research on Cross-Domain Recommender Systems (CDRS) through the development of an ontological framework that structures and clarifies the core concepts, processes, and design choices underlying effective knowledge transfer across domains. Based on a systematic review of 37 peer-reviewed studies, the proposed ontology identifies six fundamental components including domain characteristics, data and preprocessing, knowledge transfer mechanisms, methods and algorithms, evaluation and validation, and application context. The analysis reveals that knowledge transfer mechanisms, including entity-based, pattern-based, and feature-based paradigms, play a central role in addressing data sparsity and cold-start problems, with their effectiveness strongly influenced by domain similarity and data availability. The study further highlights the evolution of CDRS methodologies from traditional matrix factorization techniques to advanced deep learning and graph-based models, demonstrating how increased model expressiveness improves cross-domain representation learning under appropriate data conditions. Evaluation practices are shown to be shifting from single-metric accuracy assessments toward multi-objective frameworks that incorporate ranking quality, coverage, and user-centric measures. Building on these findings, this paper translates the proposed ontology into a practical decision framework that guides practitioners in selecting suitable transfer strategies, algorithms, and evaluation metrics based on domain characteristics and system constraints. Overall, this work contributes a unified conceptual foundation and actionable guidance for advancing CDRS research and deployment across diverse real-world application domains.