Abstract: In recent years, with the increasing abundance of high-quality data, continuous development of machine learning techniques and significant improvements of computational capabilities, data-driven inventory management is experiencing unprecedented development opportunities. However, comprehensive and systematic reviews of research advances in this emerging field are currently lacking. In this study, an in-depth analysis of 183 academic papers is conducted using bibliometrics, and the state of the art in this field is visualized through scientific knowledge graphs. Then, the research results of data-driven inventory management from the perspectives of big data and operation management are summarized and synthesized in three aspects: demand information, basic models and basic methods. Essentially, this paper introduces four inventory management models from the perspectives of demand uncertainty and feature data: univariate data-driven newsvendor model, univariate data-driven dynamic inventory model, multi-feature data-driven newsvendor model and multi-feature data-driven dynamic inventory model. On this basis, six main data-driven decision-making methods are summarized: Bayesian analysis, robust optimization, sample average approximation, quantile regression, operation statistics and machine learning. Finally, future research directions and suggestions are discussed from the perspectives of methodologies, tools, challenges, and application hotspots in data-driven inventory management, aiming to provide valuable references and insights for researchers and practitioners in the relevant fields, and to foster the continuous development of data-driven inventory management.

Key words: data-driven, inventory management, newsvendor, dynamic inventory, literature review