Natural Language Processing for Financial Regulation: FinregE innovation in collaboration with Imperial College

Natural Language Processing for Financial Regulation

In 2021, FinregE, a trusted RegTech SME, was awarded a highly competitive Smart Grant from Innovate UK to develop an innovative natural language processing (NLP) platform for financial regulation. They collaborated with world-renowned Imperial College London on this research project to unlock the potential of AI for regulatory compliance. 

 After intense research and development, FinregE is proud to announce that this collaboration has now resulted in a publication – “Natural Language Processing for Financial Regulation” published in December 2023. This paper provides key insights into how advanced NLP techniques like transformer networks and attention mechanisms can be applied to the complex legal and regulatory text that financial institutions deal with. 

 The paper explores how natural language processing (NLP) techniques can be applied to improve semantic search and matching between financial rules and policies. In the absence of large, labelled datasets, the authors relied on FinregE’s extensive database of financial policies as a corpus for unsupervised domain adaptation methods. By further pretraining sentence encoder models like BERT on this corpus, the authors can significantly improve the accuracy of identifying related rule-policy pairs compared to off-the-shelf NLP models. 

Both supervised and unsupervised techniques are explored, including a novel generative pseudo-labelling approach that automatically creates query-policy pairs from the corpus for fine-tuning. This allows domain adaptation without needing manually labelled examples, overcoming a key challenge. 

 FinregE’s database and software tools were instrumental in providing the financial policy corpus for pretraining and evaluating model performance. Without this domain-specific data, improving semantic matching accuracy would have been incredibly difficult. 

The paper demonstrates how even limited domain data, when properly utilized through transfer learning and unsupervised Adaptative Pretraining methods, can lead to better NLP systems. The improvements have important implications for financial institutions trying to manage regulatory compliance across thousands of policies. 

 The authors of the paper thanks FinregE for enabling this research through access to real-world financial policies. As NLP continues rapid progress, close collaboration between researchers and industry partners like FinregE will be key to developing and validating new techniques on complex real-world problems.

Read the paper here: Natural Language Processing for Financial Regulation by Antoine (Jack) Jacquier, Dragos Gorduza, Ixandra Achitouv :: SSRN 

Downloads Alert