To address the issues of insufficient annotated entity-relationship data and the substantial labor demands of manual annotation in Tibetan medicine texts, this study proposes an automatic annotation method based on domain-specific fine-tuning of Yak-LLaMA-7B. Entries from the Tibetan medicine dictionary were converted into instruction datasets, and Low-Rank Adaptation(LoRA) was used for efficient parameter fine-tuning. Furthermore, we incorporated a knowledge fusion layer and semantic consistency regularization to generate entity-relationship triples. The automatically annotated results were manually reviewed and filtered to create a structured dataset. Model performance evaluation experiments showed that Yak-LLaMA-7B attained an F1 score of 82.31%, representing improvements of 2.49, 5.77, and 7.18 percentage points over TiLamb-7B, Tibetan-LLaMA2-7B, and Tibetan-Alpaca-7B, respectively. Quantization comparison results show that 4-bit quantization led to a modest decrease in F1 score from 82.3% to 81.1%(a reduction of 1.2 percentage points), while concurrently reducing GPU memory consumption from 18 GB to 8 GB-an approximately 55.6% reduction. These results indicate that the method can be used for auxiliary annotation of entity-relationships in Tibetan medicine texts, providing a structured data foundation for the construction of a Tibetan medicine knowledge graph.
藏医药是中华民族传统医学的重要组成部分,其知识体系蕴含大量未结构化的古籍文献与专业术语,如药材属性、方剂配伍和病症关联。对上述知识进行系统化整理是藏医药文献研究的关键任务[1]。已有藏医药古籍数字化及标准体系建设研究[2],为文献的数字化保存和利用奠定了基础,但将数字化文本进一步转化为可计算的实体关系数据,仍需明确实体边界、统一关系类型并控制标注质量。大语言模型(Large Language Model,LLM)通过预训练学习文本表示,并借助领域微调适配具体任务[3],为上述结构化处理提供了技术路径。
PereiraC A, KomarluT, MobeirekW. The future of AI-assisted writing[PP/OL]. V1. arXiv (2023-06-29) [2026-09-08]. DOI:10.48550/arXiv.2306.16641 .
[5]
ShiS, ZhaoE, BiW, et al. Effidit: an assistant for improving writing efficiency[C]//Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics: Volume 3, System Demonstrations. Toronto: Association for Computational Linguistics, 2023:508-515.
[6]
YuanA, CoenenA, ReifE, et al. Wordcraft: story writing with large language models[C]//Proceedings of the 27th International Conference on Intelligent User Interfaces. New York: Association for Computing Machinery, 2022:841-852.
[7]
MøllerA G, PeraA, DalsgaardJ, et al. The parrot dilemma: human-labeled vs. LLM-augmented data in classification tasks[C]//Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics: Volume 2, Short Papers. St. Julian’s, Malta: Association for Computational Linguistics, 2024:179-192.
[8]
ZhaoH H, ZhouP, ShouM Z. GENIXER: empowering multimodal large language models as a powerful data generator[C]//Computer Vision-ECCV 2024. Cham: Springer, 2025:129-147.
[9]
KimJ M, BaderJ, AlanizS, et al. DataDream: few-shot guided dataset generation[C]//Computer Vision-ECCV 2024. Cham: Springer, 2025:252-268.
[10]
LiuX, JiK, FuY, et al. P-Tuning v2: prompt tuning can be comparable to fine-tuning universally across scales and tasks[PP/OL]. V3. arXiv (2022-03-20) [2026-09-08]. DOI:10.48550/arXiv.2110.07602 .
[11]
TeamG L M, ZengA, XuB, et al. ChatGLM: a family of large language models from GLM-130B to GLM-4 All Tools[PP/OL]. V2. arXiv (2024-07-30) [2026-09-08]. DOI:10.48550/arXiv.2406.12793 .
[12]
XiongH L, WangS, ZhuY T, et al. DoctorGLM: fine-tuning your Chinese doctor is not a herculean task[PP/OL]. V2. arXiv (2023-04-17) [2026-09-08]. DOI:10.48550/arXiv.2304.01097 .
[13]
ZhangH B, ChenJ Y, JiangF, et al. HuatuoGPT, towards taming language model to be a doctor[C]//Findings of the Association for Computational Linguistics: EMNLP 2023. Singapore: Association for Computational Linguistics, 2023:10859-10885.
[14]
ChenY R, WangZ Y, XingX F, et al. BianQue: balancing the questioning and suggestion ability of health LLMs with multi-turn health conversations polished by ChatGPT[PP/OL]. V2. arXiv (2023-12-04) [2026-09-08]. DOI:10.48550/arXiv.2310.15896 .
[15]
TouvronH, LavrilT, IzacardG, et al. LLaMA: open and efficient foundation language models[PP/OL]. V1. arXiv (2023-02-27) [2026-09-08]. DOI:10.48550/arXiv.2302.13971 .
[16]
WangH C, LiuC, XiN W, et al. HuaTuo: tuning LLaMA model with Chinese medical knowledge[PP/OL]. V1. arXiv (2023-04-14) [2026-09-08]. DOI:10.48550/arXiv.2304.06975 .