Part 7/12:

The core research question posed was whether pre-trained LLMs—especially those fine-tuned with domain-specific data—could effectively undertake end-to-end credit assessment tasks. Two hypotheses guided this inquiry:

1. Can pre-trained LLMs, with extensive prior knowledge, perform credit assessments?

2. Does fine-tuning with relevant, credit-specific data improve their performance?

The study involved dividing data into two categories: digital and non-traditional data (such as behavioral and contact data) and various loan types with different evaluation criteria.

Data Preparation and Prompt Engineering