Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LendNova: Towards Automated Credit Risk Assessment with Language Models

Kiarash Shamsi, Danijel Novokmet, Joshua Peters, Mao Lin Liu, Paul K Edwards, Vahab Khoshdel

TL;DR

LendNova addresses automated credit risk assessment by building an end-to-end pipeline that operates directly on raw, jargon-heavy credit bureau text using language models. It introduces the credit-story representation, a data-preparation workflow with segmentation, translation, and temporal analysis, and a parallel FinBERT-based modeling approach followed by an MLP predictor. The work demonstrates near-baseline predictive performance on real-world data while highlighting substantial cost-efficiency through reduced preprocessing, licensing, and data-engineering needs, paving the way for scalable foundation-model ciclo decisions in finance. Collectively, LendNova lays groundwork for intelligent, adaptable credit risk agents and multi-task financial foundation models that can autonomously interpret bureau data for diverse risk-management tasks.

Abstract

Credit risk assessment is essential in the financial sector, but has traditionally depended on costly feature-based models that often fail to utilize all available information in raw credit records. This paper introduces LendNova, the first practical automated end-to-end pipeline for credit risk assessment, designed to utilize all available information in raw credit records by leveraging advanced NLP techniques and language models. LendNova transforms risk modeling by operating directly on raw, jargon-heavy credit bureau text using a language model that learns task-relevant representations without manual feature engineering. By automatically capturing patterns and risk signals embedded in the text, it replaces manual preprocessing steps, reducing costs and improving scalability. Evaluation on real-world data further demonstrates its strong potential in accurate and efficient risk assessment. LendNova establishes a baseline for intelligent credit risk agents, demonstrating the feasibility of language models in this domain. It lays the groundwork for future research toward foundation systems that enable more accurate, adaptable, and automated financial decision-making.

LendNova: Towards Automated Credit Risk Assessment with Language Models

TL;DR

LendNova addresses automated credit risk assessment by building an end-to-end pipeline that operates directly on raw, jargon-heavy credit bureau text using language models. It introduces the credit-story representation, a data-preparation workflow with segmentation, translation, and temporal analysis, and a parallel FinBERT-based modeling approach followed by an MLP predictor. The work demonstrates near-baseline predictive performance on real-world data while highlighting substantial cost-efficiency through reduced preprocessing, licensing, and data-engineering needs, paving the way for scalable foundation-model ciclo decisions in finance. Collectively, LendNova lays groundwork for intelligent, adaptable credit risk agents and multi-task financial foundation models that can autonomously interpret bureau data for diverse risk-management tasks.

Abstract

Credit risk assessment is essential in the financial sector, but has traditionally depended on costly feature-based models that often fail to utilize all available information in raw credit records. This paper introduces LendNova, the first practical automated end-to-end pipeline for credit risk assessment, designed to utilize all available information in raw credit records by leveraging advanced NLP techniques and language models. LendNova transforms risk modeling by operating directly on raw, jargon-heavy credit bureau text using a language model that learns task-relevant representations without manual feature engineering. By automatically capturing patterns and risk signals embedded in the text, it replaces manual preprocessing steps, reducing costs and improving scalability. Evaluation on real-world data further demonstrates its strong potential in accurate and efficient risk assessment. LendNova establishes a baseline for intelligent credit risk agents, demonstrating the feasibility of language models in this domain. It lays the groundwork for future research toward foundation systems that enable more accurate, adaptable, and automated financial decision-making.
Paper Structure (28 sections, 3 equations, 4 figures)

This paper contains 28 sections, 3 equations, 4 figures.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Conventional Credit Risk Assessment
  4. Language Models in Finance
  5. Foundation and Automated Models in Finance
  6. Data Sources and Integration
  7. Dataset Characteristics
  8. Target Definition
  9. Ethical Consideration
  10. Data Challenges
  11. System Overview
  12. Data Preparation
  13. Record Parser Module
  14. Segmentation Module
  15. Translation Module
  16. ...and 13 more sections

Figures (4)

  • Figure 1: Potential effect of transition from bureau aggregated features to a full data usage without feature aggregation.
  • Figure 2: LendNova System Architecture. This figure shows the three main components of LendNova: (a) Data Preparation, transforming raw input into structured credit stories $St_n$ (e.g., $St_1, St_2, \dots$); (b) Language Model, embedding credit stories with temporal vectors with each tokenized segment represented as $TS_n$ (e.g., $TS_1, TS_2, \dots$); and (c) Task Predictor, training on these embeddings for final predictions.
  • Figure 3: Performance Trend of Model Versions. This chart shows the steady improvement in AUC scores across model development stages, with our final version achieving close alignment to the industry's ideal baseline, reflecting the potential of LendNova in credit risk prediction.
  • Figure 4: AUC of Champion Model Across Data Splits