https://link.springer.com/article/10.1007/s11135-026-02837-4
Artificial intelligence is now deeply embedded in finance. Banks use machine learning to assess credit risk, investors rely on algorithms to anticipate market movements, and financial institutions increasingly automate fraud detection, valuation and decision support. Yet the better these models become, the harder they can be to understand.
This problem motivated our paper, “Explainable artificial intelligence in finance: a bibliometric and topic modeling analysis using BERTopic.” We wanted to understand how research on explainable AI in finance has developed and what themes dominate the field.
Explainable artificial intelligence in finance is a highly relevant topic since in finance opaque decisions cannot be accepted casually. A loan rejection, credit rating, risk warning or ESG investment signal can have real consequences. In these settings, accuracy matters, as do transparency, accountability and trust.
When we began, we noticed that the literature was growing quickly but becoming fragmented. Many studies focused on particular applications such as credit scoring, fraud detection or market forecasting. What was missing, was a broader map of the field. We therefore combined co-word analysis with BERTopic modeling to examine 90 peer-reviewed articles selected from Scopus and Web of Science.
Th two methods we applied gave us complementary views. The co-word analysis showed the visible structure of the field: machine learning for financial risk assessment, explainable decision-making during periods of uncertainty, interpretable AI for credit scoring and digital finance, and explainable ensemble learning for macroeconomic forecasting.
BERTopic added a more semantic layer. It showed that the largest body of work focuses on credit risk and financial decision support. This was not surprising, but it was striking how dominant this area remains. Other topics were smaller but still highly relevant: asset pricing, market dynamics, portfolio risk modeling, correlation analysis, automated valuation, trading and sovereign risk forecasting.
One core question emerged: should explainability be added after a model is built, or should it be designed into financial AI systems from the start? Our findings suggest that the field is gradually moving toward the latter.
Our findings matter for practice and academia. For financial institutions, explainability should not be treated as a regulatory checkbox. It can improve model governance, reveal bias, support better communication with customers and help professionals make more confident decisions. For regulators, it offers a way to examine complex AI systems without dismissing their potential. For researchers, it opens important questions around fairness, ESG transparency, uncertainty, multimodal financial data and real-time decision support.
The study also illustrates that the field is moving quickly. Any review of AI research captures a quickly moving landscape. Still, we hope our paper provides a useful snapshot of where explainable AI in finance stands today and where it may go next.
Ultimately, the black box is not only a technical problem. It is also a governance problem, a communication problem and a trust problem. Our aim was to show how researchers are beginning to address this challenge, and why explainability is becoming a central pillar of responsible financial innovation.
The full paper can be found here:
https://link.springer.com/article/10.1007/s11135-026-02837-4
Abderahman Rejeb is a researcher at the Faculty of Business and Economics of Széchenyi István University in Győr, Hungary. His published work is situated at the intersection of business, technology, and supply chain research, with a strong interest in emerging digital applications and interdisciplinary analysis.
Karim Rejeb is affiliated with the Faculty of Sciences of Bizerte at the University of Carthage, Tunisia. His academic profile reflects research activity in information and communication technologies, with public records also linking his work to artificial intelligence, cybersecurity, and bibliometric studies.
Horst Treiblmaier is a Full Professor at Modul University Vienna, Austria. His research interests include the business implications of blockchain and artificial intelligence, the evolution of Web3, and digital transformation in general. He teaches blockchain-related topics and frequently speak at academic conferences and industry events. In 2022, he won the Blockchain Frontier Award from the Blockchain Research Institute (BRI).