组织工程与重建外科杂志 ›› 2024, Vol. 20 ›› Issue (6): 643-.

• • 上一篇    下一篇

大语言模型结合数字人技术合成短视频在医学科普中的效果评价

  

  • 出版日期:2024-12-02 发布日期:2025-01-03

Evaluation of the effectiveness of short videos synthesized by large language models combined with human video generation in medical science popularization

  • Online:2024-12-02 Published:2025-01-03

摘要:

目的 大语言模型结合数字人技术已被广泛应用于短视频的制作,本研究旨在综合评估大语言模型结合虚拟数字人技术辅助医生制作短视频科普内容的效果。方法 在短视频平台收集4种不同主题的医生讲座视频,每个主题包括1个医生数字人视频和2个真实医生拍摄的视频。向志愿者随机展示视频后发放问卷,每个视频由至少15名志愿者进行评估,志愿者并不知情视频是否由数字人技术生成。问卷评分包括真实性感知、内容质量、视听质量、专业水平、亲切度、推荐意愿和接受度7个方面。结果 共回收198份有效的评估问卷,其中数字人视频问卷66份,真实视频132份。真实拍摄视频与数字人视频的7个问题得分没有统计学差异,且与性别、收入水平和学历无关。29.9%的志愿者认为人工智能生成的视频会影响其对医生的信任,而36.95%的志愿者认为不会改变其对医生的整体看法。结论 当前的数字人技术已经能够制作出难以分辨真实性的科普短视频,有效辅助了视频质量和医生形象的呈现。今后需要加强相关法律法规的建设,以及使用加密技术、水印和其他数字技术手段,以保护个人形象不被未经授权的复制和使用,防止潜在的伦理风险。

关键词:

Abstract:

Objective The combination of large language models and human video generation technology has been widely applied in the production of short videos. This study aims to comprehensively evaluate the effectiveness of large language models combined with human video generation technology in assisting doctors to create short video content for public education. Methods Doctor lecture videos on four different topics from short video platforms were collected, each topic including one video featuring a virtual digital human doctor and two videos featuring real doctors. After randomly presenting videos of 12 doctors to volunteers, questionnaires were distributed, with each video being assessed by at least 15 volunteers who were unaware whether the videos were generated by digital human technology. The questionnaire included seven aspects: Perceived authenticity, content quality, audio-visual quality, professional level, cordiality, willingness to recommend, and acceptance. Results A total of 198 valid assessment questionnaires were collected, including 66 for virtual digital human videos and 132 for real videos. There were no statistically significant differences in the scores for the seven questions between real and virtual digital human videos, and these differences were not related to gender, income level, or educational background. 29.9% of volunteers believed that discovering the videos were generated by artificial intelligence would affect their trust in the doctors, while 36.95% of volunteers believed that the generation of videos by artificial intelligence would not change their overall perception of the doctors. Conclusion Current human video generation technology is capable of producing indistinguishable educational short videos, effectively assisting in the presentation of video quality and the image of  doctors. It is necessary to strengthen legal regulations and use encryption technology,watermarks, and other digital techniques to protect personal images from unauthorized copy and use, preventing potential ethical risks.

Key words: