Yanting Zhang

Ms. Yanting Zhang

orcid

Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
9 May 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
ZKZhizhuo Kou
Zhizhuo Kou
ZQZhiqiang Qian
Zhiqiang Qian
ZZZhenghao Zhu
Zhenghao Zhu
JXJiyuan Xin
Jiyuan Xin
YCYakun Cui
Yakun Cui
YGYike Guo
Yike Guo
Credit risk assessment requires both accurate prediction and structured decomposition of how hetero-geneous evidence contributes to each decision. Monolithic Large Language Models can incorporate unstructured evidence and natural-language reasoning into such workflows, but in high-stakes un- derwriting they may be distracted by noisy inputs, miss rare but decisive risk cues, and offer limited control over policy-dependent decision thresholds. We present CREDITAGENT, a hierarchical credit review system with three stages: evidence filtering, specialist risk analysis by agents, and decision fusion. Our central contribution is holding the adapted backbone, specialist-agent outputs, hard-stop rules, and data split fixed, we vary only the final fusion strategy to isolate the effect of hierarchical fusion on underwriting quality. On a held-out set of 6,000 personal credit cases from Chinese financial institution, CREDITAGENT achieves 83.32% accuracy and a Business Efficiency Coefficient of 0.7647 outperform flagship model. We present these findings as an institution-specific case study while identifying which components (hierarchical fusion, GRPO training recipe) are mechanism-portable versus institution-specific (hard-stop rules, cost ratios). To ensure reproducibility, we make code and dataset publicly available at https://github.com/kouzhizhuo/Credit_Agents.

21 April 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
ZKZhizhuo Kou
Zhizhuo Kou
YZYanting Zhang
Yanting Zhang
LZLei Zhu
Lei Zhu
ZZZhenghao Zhu
Zhenghao Zhu
YCYakun Cui
Yakun Cui
YGYike Guo
Yike Guo
While Large Language Models (LLMs) have shown great promise in transforming credit risk assess-ment, existing evaluation frameworks are almost exclusively concerned with general financial NLP tasks and neglect the specific reasoning needed by practitioners. To address this, we develop the Credit Context Log Understanding and Prediction Evaluation (CCLUPE) benchmark. Unlike the previous benchmarks, CCLUPE aims to capture and evaluate the intricate reasoning unique to each constituent of the Chinese credit market, where evaluations are heavily based on the integration and synthesis of complex transacted logs and the prediction of hidden financial behaviors. Unlike previous benchmarks, CCLUPE aims to capture and evaluate the intricate reasoning unique to each constituent of the Chinese credit market. Unlike previous benchmarks, CCLUPE aims to capture and evaluate the intricate reasoning unique to each constituent. CCLUPE boasts more than 4,000 premium samples segmented by individual and micro-enterprise customers and distributed among 7 principal log types and 16 sub log types. A comprehensive assessment process involving upwards of 20 professional annotators is enacted to guarantee the quality of the dataset. Moreover, we introduce Log-Score, a novel evaluation metric designed to incorporate log misunderstanding penalties and assess multifaceted competencies. Even the state-of-the-art models underperform when it comes to these high-stakes tasks. CCLUPE serves as a rigorous testbed for the next generation of financial LLMs, ensuring their robustness for deployment in complex real credit scenarios.