Review on the Use of Machine Learning and Artificial Intelligence in Accounting Research

Abstract

This study (1) systematically analyzes and compares the current state of accounting research utilizing machine learning and artificial intelligence (ML/AI) technologies between international and domestic contexts from 2015 to 2025, (2) diagnoses the current position of domestic research, and (3) propose future development directions. The rapid advancement of ML/AI technologies has created unprecedented opportunities for accounting research to expand beyond traditional methodological boundaries, enabling researchers to address previously intractable questions and develop novel theoretical insights. Against this backdrop, understanding the current landscape and identifying research gaps becomes crucial for advancing the field. To achieve these objectives, we conducted a comprehensive literature review by selecting 61 papers from major international accounting journals and 13 papers from Korean Citation Index (KCI) registered journals. Our analysis employed a multidimensional classification framework consisting of three key dimensions: ML/AI utilization types, research topics, and data types. The ML/AI utilization types were further categorized into four distinct areas: predictive research (forecasting future outcomes), diagnostic research (quantifying previously unmeasurable concepts), descriptive research (discovering new phenomena through data exploration), and prescriptive research (developing practical solutions and recommendations). This comprehensive framework allows for a nuanced understanding of how ML/AI technologies are being integrated into accounting research across different methodological approaches. We find that international research has experienced dramatic growth since 2022, demonstrating a balanced development between predictive (37.7%), measurement (34.4%), descriptive (13.1%) and prescriptive (13.1%) research. The measurement research domain has achieved particularly innovative outcomes by quantifying previously unmeasurable concepts such as earnings virality, CEO depression, and voice delivery quality, thereby expanding the theoretical horizons of accounting research. These developments represent a shift in accounting research methodology, moving beyond traditional archival approaches to embrace data-driven, inductive methodologies that can uncover hidden patterns and relationships in financial data. By research topics, the development has been evenly distributed across auditing (27.9%), capital market research (24.6%), and financial reporting (24.6%), indicating a broad-based adoption of ML/AI technologies across different accounting subfields. In terms of data utilization, researchers have diversely employed structured financial data (47.5%), textual data (21.3%), and alternative data (18.0%), showcasing the versatility of ML/AI applications in processing various data formats and sources. In stark contrast, domestic research exhibits significant temporal and qualitative gaps compared to international counterparts. Korean accounting research in this domain began 3-4 years later than international research and shows substantial disparities in both scale and quality. Domestic research demonstrates a pronounced concentration on prediction-focused studies (61.5%) and reliance on structured financial data (76.9%), resulting in limited research scope and methodological approaches. More critically, while international research extends theoretical boundaries through the development of novel measurement indicators and creative research designs, domestic research primarily remains confined to performance verification of methodologies. This gap reflects deeper structural challenges including limited access to alternative data sources, insufficient interdisciplinary collaboration, or inadequate computational resources. The academic contributions of this study are threefold. First, this research represents the first systematic literature review of ML/AI-related accounting research, providing an objective assessment of the current status. By establishing a comprehensive baseline understanding, this study serves as a foundation for future research and policy development in this rapidly evolving field. Second, through our multidimensional classification framework encompassing utilization types, research topics, and data types, we analyze the multilayered impact of ML/AI technologies on accounting research. This framework provides a structured approach for understanding the diverse ways in which ML/AI can contribute to accounting knowledge and practice. Third, through systematic comparison between international and domestic research, we specifically identify the current position of domestic research and concrete development challenges, providing actionable insights for improving the research landscape. Our findings reveal that ML/AI technologies are driving a paradigm shift in accounting research methodology, enabling the quantification of previously unmeasurable concepts and suggesting the possibility of supplementally utilizing data-driven inductive approaches to discover new theoretical insights. This methodological evolution has important implications for how accounting researchers approach their work, moving from purely theory-driven deductive approaches to embracing the potential of data-driven discovery. Finally, this study proposes that future research development requires advancement across four key areas corresponding to our classification framework. In predictive research, there is a need for theory-based research designs that go beyond simple performance improvement to provide meaningful accounting insights. In measurement research, the quantification of domestically unique concepts that differentiate from international precedents could make significant contributions to accounting scholarship in the Korean context. In explanatory research, the utilization of alternative data to discover new phenomena represents a promising avenue for expanding accounting knowledge. In prescriptive research, the development of systematic educational methodologies is essential for building ML/AI capabilities within the accounting profession. Through these multifaceted approaches, we expect that domestic accounting research can play a more distinctive and leading role in the global ML/AI accounting research ecosystem, ultimately contributing to both theoretical advancement and practical improvement in accounting practice.