目的 基于自然语言处理和深度学习语义建模方法，分析黄芪文献的研究现状、主题结构及演化特征，为黄芪研究热点识别和知识结构梳理提供参考。方法 检索PubMed、Web of Science、Embase、中国知网、万方数据和维普资讯数据库中黄芪相关文献，检索字段限定为题名和关键词。经NoteExpress去重及人工筛选后，以题名/标题、关键词和摘要构建联合文本，采用文本清洗、术语规范化和语义嵌入等自然语言处理方法，结合PubMedBERT与BERTopic模型进行主题识别、可视化分析和Topic 0子主题分析。结果 最终纳入中文文献12 159篇，英文文献1 536篇。中文文献2015年后发文量明显增加，英文文献总体呈持续增长趋势。主题建模结果显示，中文文献形成19个主要主题，主要涉及黄芪免疫调节、肿瘤免疫调控、改善周围神经病变、糖尿病肾病防治等药理作用，以及黄芪甲苷质量控制和磷脂酰肌醇-3-羟激酶（phosphatidylinositol-3-hydroxykinase，PI3K）-蛋白激酶B（protein kinase B，Akt）通路机制等方向；英文文献形成20个主要主题，主要涉及黄芪方药机制、质量分析、化学成分以及与癌症、纤维化、缺血再灌注损伤、糖尿病及肠道微生态相关的研究方向。Topic 0子主题分析显示，黄芪相关中文文献核心主题呈现“成分-制剂-复方-疾病-机制”交织特征，英文文献核心主题更突出机制化和模型化表达。结论 黄芪的中英文研究均呈现持续增长和多主题分化趋势，中文文献更侧重临床应用、复方制剂、糖尿病肾病和质量控制，英文文献更侧重分子机制、网络药理学、天然产物化学和现代药理验证。自然语言处理与深度学习主题建模方法可有效揭示黄芪研究的知识结构和演化路径，为后续研究选题和证据整合提供参考。

Objective To analyze the research status, thematic structure and evolutionary trends of Chinese and English literature on Huangqi (Astragali Radix) using natural language processing and deep-learning-based semantic modeling, and to provide references for hotspot identification and knowledge structure mapping. Methods Literature related to Astragali Radix was retrieved from PubMed, Web of Science, Embase, China National Knowledge Infrastructure (CNKI), Wanfang Data and VIP databases, with the retrieval fields limited to titles and keywords. After deduplication using NoteExpress and manual screening, titles, keywords and abstracts were combined to construct the analytical corpus. Natural language processing approaches including text cleaning, terminology normalization and semantic embedding were performed, and PubMedBERT combined with BERTopic was used for topic identification, visualization and subtopic analysis of Topic 0. Results A total of 12 159 Chinese publications and 1 536 English publications were included. The publication output of Chinese publications increased markedly after 2015, while English publications showed a continuous upward trend. Topic modeling identified 19 major topics in Chinese literature, mainly involving immunomodulation, diabetic nephropathy, quality control of astragaloside IV, peripheral neuropathy, tumor immunoregulation and phosphatidylinositol-3-hydroxykinase (PI3K)-protein kinase B (Akt)-related network pharmacology. In English literature, 20 major topics were identified, mainly involving formula-related mechanisms, quality analysis of Astragali Radix, network pharmacology and cancer, chemical constituents, fibrosis, ischemia-reperfusion injury, diabetes and gut microbiota. Subtopic analysis of Topic 0 suggested that the Chinese core topic showed an intertwined pattern of “components-preparations-formulas-diseases-mechanisms”, whereas the English core topic was more mechanism- and model-oriented. Conclusion Both Chinese and English Astragali Radix literature showed continuous growth and thematic diversification. Chinese studies focused more on clinical application, compound formulas, diabetic nephropathy and quality control, whereas English studies emphasized molecular mechanisms, network pharmacology, natural product chemistry and modern pharmacological validation. Natural language processing and deep-learning-based topic modeling can effectively reveal the knowledge structure and evolutionary trajectory of Astragali Radix research, providing references for subsequent research topic selection and evidence integration.

TP18;G350;R282.71

广州市科技局基础研究计划（2023A03J0306）；广州市科学技术局青年博士“启航”项目（2025A04J3795）；2026年广州中医药大学“揭榜挂帅”青年拔尖人才团队项目（郑文江）