| 摘要: |
| 随着以大模型为代表的新一代人工智能在消费金融领域的日益普及，其应用场景已从机构问答、权益推荐、文案生成、用户理解等相对简单的任务逐渐扩展到篡改识别、法条适用、风控识别、系统链路分析和财经分析等复杂任务。提出一套面向金融行业的大模型可信应用框架，旨在应对这些模型在专业性、可控性、真实性与安全性方面面临的挑战。该框架采用一系列技术措施来确保上述目标的实现，包括但不限于优化数据供给质量、强化模型在特定应用领域内的专业知识能力、提升生成内容的可控性和可靠性、增强智能体推理过程中的信任度、运用围栏机制和安全防护策略、构建全面且详尽的评估体系，以及形成“反馈－迭代”良性循环机制。通过在交互式信用成长、智能营销、机构服务和消保纠纷化解等多个实际场景中的应用案例研究，不仅验证了可信框架的有效性、安全性及其准确性，也证明了该框架对业务效果的显著提升。 |
| 关键词: 关键词：金融大模型；大模型可信应用；数字化转型 |
| DOI： |
| 分类号:F830.5 |
| 基金项目: |
|
| Large Model Trustworthy Applications and Practices in the Consumer Finance Sector |
|
ZHAO Jie,YANG Sen,WANG Yi,WANG Wei,CHEN Bin
|
|
(Chongqing Ant Consumer Finance Co.,Ltd)
|
| Abstract: |
| With the widespread application of large models,representing the new generation of artificial intelligence in consumer finance,their use cases have expanded from simple tasks such as institutional Q&A,benefits recommendation,content creation,and user understanding to more complex tasks like tampering detection,legal clause application,risk control identification,and system link analysis.In response to this trend,this paper introduces a trustable application framework for large models in the financial sector,aimed at addressing issues of professionalism, controllability, authenticity,and security.The framework employs a series of technical measures to ensure these aspects,including improving data supply quality,enhancing the model’s professional capabilities in its application domain,increasing the controllability and credibility of the generated content,boosting the trustworthiness of intelligent agent reasoning,using fencing tools and security measures,establishing a comprehensive and sufficient evaluation system,and building a “feedback-iteration” virtuous cycle system.These measures significantly enhance the trustworthiness of large model applications.Through practical applications in interactive credit growth,intelligent marketing,institutional services,and consumer protection dispute resolution,the effectiveness,security,and accuracy of this trustable framework have been verified,effectively improving business outcomes. |
| Key words: Key words：Financial Large Model;Trustworthy Application of Large Models;Digital Transformation |
