Enhancing Information Processing Capabilities in Financial Services: Utilizing Machine Learning Techniques to Overcome Limited Data Availability

Abstract

This thesis presents an investigation into the application of advanced machine learning techniques to enhance information processing capabilities in the financial services sector, focusing on the challenges faced by small and medium-sized enterprises (SMEs) due to limited data availability. It comprises three interconnected studies that utilize cutting-edge approaches like graph neural networks and large language models to overcome the specific challenges of financial analytics and risk assessment in SMEs.

In the first study, the thesis introduces a relational graph attention network algorithm. This methodological innovation addresses limited data availability in SMEs by utilizing information from adjacent enterprises. It constructs a network that factors in shared management teams and business interactions, aiming to refine the accuracy of credit risk predictions. This approach is empirically validated using a dataset from Chinese SMEs, demonstrating its potential effectiveness in practical applications.

The second study advances this line of inquiry by developing a heterogeneous spatial-temporal graph convolutional network. This model goes beyond previous approaches by integrating dynamic relational and transactional data from a network of enterprises. The study tests the model's capability to enhance the accuracy of credit default predictions, comparing its performance against traditional models that focus primarily on individual SME attributes. The empirical analysis, grounded in robust statistical evaluation, confirms the model's effectiveness.

The third study of the thesis explores a novel text mining workflow that incorporates large language models, like ChatGPT, along with an automated web crawling technique. This approach aims to extract meaningful insights from internet-sourced textual data, an important resource in scenarios of limited data availability. The practical utility of this method is showcased through a case study on the Chinese market for new energy buses, illustrating its potential application in financial service analytics.

Overall, this thesis contributes to the academic discourse in financial services by thoroughly exploring the use of machine learning methods for tackling data analysis and risk assessment challenges in SMEs with limited data. It underscores the transformative impact of machine learning in finance, offering more advanced tools for decision-making to both service providers and policymakers. Furthermore, this research lays a strong groundwork for further studies in this area and stands as a useful reference for both scholars and industry professionals.