面向中医药大模型的知识增强方法研究

doi:10.3778/j.issn.1673-9418.2407082

摘要/Abstract

摘要： 近年来，大语言模型（LLM）在各个领域取得了许多重大成果。由于缺乏专业知识，以及中医和现代医学的思想不同，大模型在中医药领域的应用仍是一项挑战。现有的知识增强方法难以保持中医方剂具有的自身结构性。为了解决以上问题，提出了一种新的知识增强方法。该方法由模型训练、图谱构建和知识增强三部分组成。在模型训练阶段，通过对基础大模型在中医药数据集上进行预训练和微调两阶段训练，得到中医药领域大模型。在图谱构建阶段，基于中医十万首经典方剂和古籍中的方剂，利用清洗后的数据集构建中医药图谱。在知识增强阶段，基于对知识图谱上信息的计算，利用检索图谱中的专业知识和图谱结构计算检索结果，中医药方剂中的结构特性得以保留。在中医药方剂配伍任务上，针对于任务特性提出了一组评价标准，包括主观指标和客观指标，用于评估模型在该任务上的表现。实验表明，该方法相对于基准测试模型，在主观指标和客观指标上均获得了较大提升，BLEU-1最高提升0.09，ROUGE-1最高提升0.21。消融实验表明，该方法对于模型在该任务上具有较大作用，未使用知识增强的模型BLEU-1相比于使用知识增强下降约37%。

关键词: 大语言模型（LLM）, 中医药, 方剂优化, 检索增强生成

Abstract: Recently, large language models (LLM) have made significant achievements in various fields. However, due to lack of specialized knowledge and the gap between modern medicine and traditional Chinese medicine (TCM), it is still a challenge to deploy LLM in TCM. Existing methods fail to maintain the structure of TCM prescription. To address the problems, a pattern of knowledge augmentation is proposed. The method includes model training, knowledge graph construction and knowledge augmentation. In the training phase, TCM language model is trained on TCM corpus, by a two-stage method combining pre-training and fine-tuning. In the knowledge graph construction phase, prescription knowledge graph is constructed from nearly 100000 preprocessed classical TCM prescriptions and those from ancient books. In the knowledge augmentation phase, enhanced by the above pattern, outputs are generated from computation of knowledge graph, according to the schema of knowledge graph from searching result, which preserves the structure of prescriptions. A set of evaluations specific to prescription optimizations is proposed, including objective and subjective indicators, to evaluate the performance of the model for the task. Experiment shows that the model improves greatly on both subjective and objective evaluations compared with baselines. BLEU-1 is increased by up to 0.09, while ROUGE-1 is increased by up to 0.21. Ablation study shows that, it is of vital importance for the model performance to be knowledge-augmented. BLEU-1 of augmentation-free model is decreased by about 37% compared with that of the augmented model.

Key words: large language model (LLM), traditional Chinese medicine, prescription optimization, retrieval augmented generation

吉祥宇, 王鑫, 张鹤译, 孟昭鹏, 张俊华, 庄朋伟, 贾勇哲, 徐大为. 面向中医药大模型的知识增强方法研究[J]. 计算机科学与探索, 2024, 18(10): 2616-2629.

JI Xiangyu, WANG Xin, ZHANG Heyi, MENG Zhaopeng, ZHANG Junhua, ZHUANG Pengwei, JIA Yongzhe, XU Dawei. Knowledge Augmentation on Traditional Chinese Medicine Language Model[J]. Journal of Frontiers of Computer Science and Technology, 2024, 18(10): 2616-2629.

参考文献

[1] TB OpenAI. ChatGPT: optimizing language models for dialogue[R]. 2022.
[2] TOUVRON H, LAVRIL T, IZACARD G, et al. LLaMA: open and efficient foundation language models[EB/OL]. (2023-02-27) [2024-05-08]. https://arxiv.org/abs/2302.13971.
[3] YANG A Y, XIAO B, WANG B N, et al. Baichuan 2: open large-scale language models[EB/OL]. (2023-09-20). [2024-05-08]. https://arxiv.org/abs/2309.10305.
[4] 王子怡, 王鑫, 张岱岩, 等. 中医药网络药理学：《指南》引领下的新时代发展[J]. 中国中药杂志, 2022, 47(1): 7-17.
WANG Z Y, WANG X, ZHANG D Y, et al. Traditional Chinese medicine network pharmacology: development in new era under guidance of network pharmacology evaluation method guidance[J]. China Journal of Chinese Materia Medica, 2022, 47(1): 7-17.
[5] 任艳, 邓燕君, 马焓彬, 等. 网络药理学在中药领域的研究进展及面临的挑战[J]. 中草药, 2020, 51(18): 4789-4797.
REN Y, DENG Y J, MA H B, et al. Research progress and challenges of network pharmacology in field of traditional Chinese medicine[J]. Chinese Traditional and Herbal Drugs, 2020, 51(18): 4789-4797.
[6] LE SCAO T, FAN A, AKIKI C, et al. BLOOM: a 176B-parameter open-access multilingual language model[EB/OL]. (2023-01-27) [2024-05-08]. https://arxiv.org/abs/2211.05100.
[7] TAORI R, GULRAJANI I, ZHANG T Y, et al. Alpaca: a strong, replicable instruction-following model[EB/OL]. (2023-03-13)[2024-05-08]. https://crfm.stanford.edu/2023/03/13/alpaca.html.
[8] CHIANG W L, LI Z H, LIN Z, et al. Vicuna: an open-source chatbot impressing GPT-4 with 90% ChatGPT quality[EB/OL]. (2023-03-30)[2024-05-08]. https://lmsys.org/blog/2023-03-30-vicuna.
[9] CUI Y M, YANG Z Q, YAO X. Efficient and effective text encoding for Chinese LLaMA and Alpaca[EB/OL]. (2024-02-23)[2024-05-08]. https://arxiv.org/abs/2304.08177.
[10] GAN R Y, WU Z W, SUN R L, et al. Ziya2: data-centric learning is all LLMs need[EB/OL]. (2024-04-04)[2024-05-08]. https://arxiv.org/abs/2311.03301.
[11] HAN T Y, ADAMS L C, PAPAIOANNOU J M, et al. Med-Alpaca: an open-source collection of medical conversational AI models and training data[EB/OL]. (2023-10-04)[2024-05-08]. https://arxiv.org/abs/2304.08247.
[12] SINGHAL K, TU T, GOTTWEIS J, et al. Towards expert-level medical question answering with large language models[EB/OL]. (2023-05-16) [2024-05-08]. https://arxiv.org/abs/2305.09617.
[13] WANG H C, LIU C, XI N W, et al. HuaTuo: tuning LLaMA model with Chinese medical knowledge[EB/OL]. (2023-04-14) [2024-05-08]. https://arxiv.org/abs/2304.06975.
[14] RAMAMURTHY R, AMMANABROLU P, BRANTLEY K, et al. Is reinforcement learning (not) for natural language processing: benchmarks, baselines, and building blocks for natural language policy optimization[C]//Proceedings of the 11th International Conference on Learning Representations, Kigali, May 1-5, 2023: 1-61.
[15] LIU J L, WANG Z M, YE Q C, et al. Qilin-Med-VL: towards Chinese large vision-language model for general healthcare[EB/OL]. (2023-11-01) [2024-05-08]. https://arxiv.org/abs/2310.17956.
[16] 崔唐明, 孙美玲, 孙华君, 等. 基于PICO模型的中医药循证指南知识图谱构建与智能问答系统研究[J]. 中国数字医学, 2024, 19(5): 20-27.
CUI T M, SUN M L, SUN H J, et al. Construction of a knowledge graph and intelligent Q&A system for evidence-based TCM guidelines based on the PICO model[J]. China Digital Medicine, 2024, 19(5): 20-27.
[17] 王嘉欣, 姚鉴玲, 马嘉慕, 等. 基于AHP-SOM聚类-TOPSIS和中医传承辅助平台研究中医药治疗围绝经期抑郁症组方规律[J]. 中草药, 2022, 53(22): 7153-7163.
WANG J X, YAO J L, MA J M, et al. Analysis on prescription rules of traditional Chinese medicine in treatment of perimenopausal depression based on AHP-SOM-TOPSIS algorithm and traditional Chinese medicine inheritance platform system[J]. Chinese Traditional and Herbal Drugs, 2022, 53(22): 7153-7163.
[18] 李德琳, 魏本征, 张诏, 等. 基于FP-growth算法的中医抗病毒方剂配伍规律探索[J]. 中华中医药学刊, 2018, 36(3): 663-668.
LI D L, WEI B Z, ZHANG Z, et al. FP-growth algorithm-based exploratory study on compatibility law of Chinese traditional anti-virus medicine prescription[J]. Chinese Archives of Traditional Chinese Medicine, 2018, 36(3): 663-668.
[19] 单雨濛, 张科, 胡文军, 等. 基于超图的中药方剂超网络中药材群组信息挖掘[J]. 中草药, 2024, 55(11): 3816-3824.
SHAN Y M, ZHANG K, HU W J, et al. Mining of group information of medicinal materials in traditional Chinese medicine prescriptions hypernetwork based on hypergraph[J]. Chinese Traditional and Herbal Drugs, 2024, 55(11): 3816-3824.
[20] 2020MEAI. TCMLLM[EB/OL]. [2024-05-08]. https://github.com/2020MEAI/TCMLLM.
[21] LEWIS P, PEREZ E, PIKTUS A, et al. Retrieval-augmented generation for knowledge-intensive NLP tasks[C]//Advances in Neural Information Processing Systems 33, Dec 6-12, 2020. Cambridge: MIT Press, 2020: 9459-9474.
[22] GUU K, LEE K, TUNG Z, et al. REALM: retrieval-augmented language model pre-training[C]//Proceedings of the 37th International Conference on Machine Learning, Jul 13-18, 2020. New York: ACM, 2020: 3929-3938.
[23] SARTHI P, ABDULLAH S, TULI A, et al. RAPTOR: recursive abstractive processing for tree-organized retrieval[C]// Proceedings of the 12th International Conference on Learning Representations, Vienna, May 7-11, 2024: 1-23.
[24] EDGE D, TRINH H, CHENG N, et al. From local to global: a graph RAG approach to query-focused summarization[EB/OL]. (2024-04-24]) [2024-05-08]. https://arxiv.org/abs/2404. 16130.
[25] SHUMAILOV I, SHUMAYLOV Z, ZHAO Y R, et al. The curse of recursion: training on generated data makes models forget[EB/OL]. (2024-04-14) [2024-05-08]. https://arxiv.org/abs/2305.17493.
[26] GUDIBANDE A, WALLACE E, SNELL C, et al. The false promise of imitating proprietary LLMs[EB/OL]. (2023-05-25) [2024-05-08]. http://arxiv.org/abs/2305.15717.
[27] GAO Y F, XIONG Y, GAO X Y, et al. Retrieval-augmented generation for large language models: a survey[EB/OL]. (2024-03-27) [2024-05-08]. https://arxiv.org/abs/2312.10997.
[28] ZHANG H B, CHEN J Y, JIANG F, et al. HuatuoGPT, towards taming language model to be a doctor[C]//Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, Singapore, Dec 6-10, 2023. Stroudsburg: ACL, 2023: 10859-10885.
[29] YANG S H, ZHAO H J, ZHU S B, et al. Zhongjing: enhancing the Chinese medical capabilities of large language model through expert feedback and real-world multi-turn dialogue[C]//Proceedings of the 38th AAAI Conference on Artificial Intelligence, Vancouver, Feb 20-27, 2024. Menlo Park: AAAI, 2024: 19368-19376.
[30] ES S, JAMES J, ESPINOSA-ANKE L, et al. RAGAs: automated evaluation of retrieval augmented generation[C]// Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, Dubrovnik, May 2-6, 2023. Stroudsburg: ACL, 2023: 150-158.