摘要：

目的 探讨2种AI工具（华知大模型和DeepSeek）在医学期刊选题策划中的应用效果，为医学期刊选题策划实际工作中有效利用AI工具提供参考和依据。方法 依据《中国实用内科杂志》办刊宗旨及《专题笔谈》栏目载文特征，以呼吸内科、消化内科和心血管内科为学科方向，利用华知大模型和DeepSeek 2种AI工具进行《中国实用内科杂志》2026年《专题笔谈》栏目的选题策划。采用“人机协同”模式，首先建立“AI工具生成-专家评估筛选-责任编辑决策”具体的“人机协同”策略并实施，然后分析AI工具的应用效果。结果 2种AI工具共生成180个选题，华知大模型和DeepSeek各90个。其中呼吸内科、消化内科和心血管内科各60题。专家评估结果显示，2种AI工具均能成功选出具有逻辑性、准确性和契合性的正确选题，但各维度及学科正确率上存在差异。进一步责任编辑评估结果显示，共筛选出有效选题55题，其中华知大模型29题，DeepSeek 26题；无效选题57题，其中华知大模型16题，DeepSeek 41题。华知大模型有效选题占比高于DeepSeek，差异有统计学意义（χ²=7.080，P=0.008）。在学科分类比较中发现，呼吸内科的有效选题占比华知大模型高于DeepSeek，差异有统计学意义（χ²=7.131，P=0.008），而消化、心血管内科的有效选题占比2种AI工具比较差异无统计学意义（χ²=0.609、1.624，P=0.435、0.202）。结论 “人机协同”模式下2种AI工具（华知大模型与DeepSeek）均可快速完成医学期刊的选题策划。华知大模型与DeepSeek展现出差异化的应用效果；共同使用可提升选题策划的效率与质量。

关键词: 医学期刊, 选题策划, 华知大模型, DeepSeek

Abstract:

Purposes This study explores the application effects of two AI tools， Huazhi large model and DeepSeek， in the topic planning of medical journals， aiming to provide references and a basis for the effective utilization of AI tools in the practical work of medical journal topic planning. Methods Based on the publication purpose of the Chinese Journal of Practical Internal Medicine and the publication characteristics of its “Topic Review” column， and focusing on the disciplines of respiratory medicine， gastroenterology， and cardiovascular medicine， two AI tools，Huazhi large model and DeepSeek， were used to plan the topics for the 2026“Topic Review” column of the journal. Under the human-AI collaboration mode， a specific human-AI collaboration strategy of“AI tool generation-expert evaluation and screening-responsible editor decision-making”was first established and implemented. Subsequently， the application efficacy of the AI tools was analyzed. Findings A total of 180 topics were generated by the two AI tools， with 90 from Huazhi large model and 90 from DeepSeek. Among these， 60 topics were generated for each of the three disciplines：respiratory medicine， gastroenterology， and cardiovascular medicine. The expert evaluation results showed that both AI tools successfully selected correct topics that demonstrated logicality， accuracy， and consistency， though differences were observed across various dimensions and disciplines in terms of accuracy rates. Further evaluation by responsible editors showed that a total of 55 valid topics were identified， including 29 generated by the Huazhi large model and 26 by DeepSeek. There were 57 invalid topics in total， among which 16 were from the Huazhi Large Model and 41 from DeepSeek. The proportion of valid topics produced by the Huazhi large model was significantly higher than that of DeepSeek， with a statistically significant difference （χ2=7.080， P=0.008）. Stratified comparison by medical discipline demonstrated that the Huazhi large model generated a higher proportion of valid topics for respiratory medicine than DeepSeek， with a statistically significant difference （χ2=7.131， P=0.008）. No statistically significant differences in the proportion of valid topics were detected between the two AI tools for gastroenterology and cardiovascular medicine （χ2=0.609， P=0.435； χ2=1.624， P=0.202）. Conclusions Under the human-AI collaboration mode， both AI tools （Huazhi large model and DeepSeek） can efficiently complete topic planning for medical journals. Huazhi large model and DeepSeek exhibit differentiated application efficacy， and their complementary use can enhance both the efficiency and quality of topic planning.

Key words: Medical journals, Topic selection planning, Huazhi large model, DeepSeek