Article Details

Abstract

To address the source conflicts and evidence-boundary problems caused by the coexistence of structured business data, institution-specific policy texts and general requests in university question answering, this study designs a university question-answering system that integrates structured queries and RAG. The system performs cascade routing according to safety constraints, structured intents, institutional knowledge and general assistance, and invokes the controlled MySQL service, the character bigram BM25 retrieval-augmentation module and the general model interface accordingly. A knowledge base is built with 30 source documents and 126 knowledge chunks, and evaluation is conducted through retrieval, routing and human assessment. The results show that the complete retrieval method achieves a Hit@3 of 97.20% and an MRR@3 of 0.863. The F1 score for institutional-knowledge routing is 0.831, the mean blind-review score of end-to-end answers is 4.24, and the mean independent review score of structured queries is 4.95. These results indicate that routing by fact source with evidence gating can improve the traceability and usability of heterogeneous university question answering.

Keywords

university intelligent question answering; institutional knowledge; RAG; hierarchical routing; BM25

References

1. [1] Lewis P，Perez E，Piktus A，et al. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks [PP/OL].arXiv:2005.11401 [cs.CL].https://doi.org/10.48550/arXiv.2005.11401.
2. [2] Karpukhin V，Oguz B，Min S，et al. Dense Passage Retrieval for Open-Domain Question Answering [PP/OL].arXiv:2004.04906 [cs.CL].https://doi.org/10.48550/arXiv.2004.04906.
3. [3] Gao Y，Xiong Y，Gao X，et al. Retrieval-Augmented Generation for Large Language Models: A Survey [PP/OL].arXiv:2312.10997 [cs.CL].https://arxiv.org/abs/2312.10997.
4. [4] Chen W H，Zha H W，Chen Z Y，et al. HybridQA: A Dataset of Multi-Hop Question Answering over Tabular and Textual Data [PP/OL].arXiv:2004.07347 [cs.CL].https://doi.org/10.48550/arXiv.2004.07347.
5. [5] Zhang Chunhong, Du Longfei, Zhu Xinning, etc Research on Educational Question Answering System Based on Large Language Model [J]. Journal of Beijing University of Posts and Telecommunications (Social Sciences Edition), 2023, 25 (6): 79-88
6. [6] Tang Bowen, Ma Mingxuan, Zhang Yining, etc Campus Q&A System Based on Intent Recognition and Retrieval Enhancement [J]. Journal of Communications, 2024, 45 (S2): 255-261
7. [7] Zhao Yan, Cheng Kai Construction of a Regulatory Q&A Large Language Model Based on Retrieval Enhanced Generation Technology [J]. Computer Applications and Software, 2025, 42 (7): 192-198
8. [8] Zhang Li'an, Li Yuchun, Chen Gongxing, etc Research on the Application of Logistics Knowledge Q&A System Based on GraphRAG [J]. Modern Information Technology, 2025, 9 (20): 40-43
9. [9] Weng Jiayi, Zhai Jie, Xiong Hanrui, etc Intelligent Q&A System for Ancient Poetry Based on Knowledge Graph and RAG Technology [J]. Modern Information Technology, 2025, 9 (20): 115-120125
10. [10] Li Chao, Jin Huifeng, Fu Hongzi, etc Research on Teaching Intelligent Question Answering System Based on DeepSeek and RAG Technology [J]. Modern Information Technology, 2025, 9 (24): 45-50
11. [11] Yi Yunheng, Pan Ji Design of Education and Teaching Knowledge Q&A System Based on Large Language Model [J]. Modern Information Technology, 2025, 9 (2): 189-194
12. [12] Robertson S，Zaragoza H. The Probabilistic Relevance Framework: BM25 and Beyond [J].Foundations and Trends in Information Retrieval，2009，4（1-2）：1-174.
13. [13] DEEPSEEK-AI. DeepSeek-V3 Technical Report [PP/OL].arXiv:2412.19437 [cs.CL].https://arxiv.org/abs/2412.19437.
14. [14] Yu H，Gan A R，Zhang K，et al. Evaluation of Retrieval-Augmented Generation: A Survey [C]//Big Data，2025：102-120.

Funding

Teaching Reform Research Project of Hunan University of International Economics (Document No. [2025] 18 of Hunan University of International Economics: No. 102); Scientific Research Project of Hunan University of International Economics (Document No. [2024] 4 of Hunan University of International Economics: No. 17)

About the authors

Gao Lei (1997-), male, Han ethnicity, from Changsha, Hunan, teaching assistant, master's student, research interests include AI applications and intelligent information systems; Corresponding author: Zhang Wanqing (1997-), female, Han ethnicity, from Luoyang, Henan, intermediate engineer, master's student, research interests include information system integration and data analysis and decision making.