The de-identified data and analysis materials
Abstract
University financial policies are often distributed across lengthy and heterogeneous documents, making it difficult for staff and students to identify complete and applicable policy provisions. This study investigates whether grounding large language model responses in retrieved institutional policy evidence can improve the completeness and traceability of university financial question answering. We developed an application-oriented policy-grounded retrieval-augmented generation system that combines policy-segment retrieval with constrained response generation. The system was evaluated on 100 Chinese financial-policy questions using a corpus of 190 policy segments and compared with Keyword Search, BM25, and a Native LLM without retrieved policy evidence. The proposed system achieved a key-point coverage of 0.824, which was significantly higher than that of all three baselines. Its traceability score was 0.840 and was significantly higher than those of Keyword Search and BM25. Correctness reached 1.62/2 and was higher than that of both retrieval baselines, although the difference from the Native LLM was not statistically significant. The proposed system also showed the lowest observed hallucination rate (4%) and over-refusal rate (2%); however, the overall hallucination difference was not statistically significant. These findings indicate that policy retrieval and constrained generation can improve answer completeness and make supporting provisions easier to locate in the…
Read the paper
Related studies
- Knowledge Management Perspective of Generative Artificial Intelligence
- Navigating Generative Artificial Intelligence Promises and Perils for Knowledge and Creative Work
- Characterizing the evolving research on enterprise content management
- Large language models and multimodal foundation models for precision oncology
- From Generative AI to Extended Reality: Multidisciplinary Perspectives on the Challenges, Opportunities, and Future of Educational Computing
- Generative Artificial Intelligence: A New Engine for Advancing Environmental Science and Engineering
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.