文章摘要
杨怡,万晨硕,李瑾,胡泽文.基于在线评论的国产生成式人工智能产品用户满意度分析[J].数字图书馆论坛,2025,21(11):62~72
基于在线评论的国产生成式人工智能产品用户满意度分析
Analysis of User Satisfaction for Domestic GenAI Products Based on Online Reviews
投稿时间:2025-10-09  
DOI:10.3772/j.issn.1673-2286.2025.11.007
中文关键词: 生成式人工智能;用户满意度;产品属性;细粒度情感分析;机器学习;重要性-绩效分析
英文关键词: GenAI; User Satisfaction; Product Attribute; Fine-Grained Sentiment Analysis; Machine Learning; IPA
基金项目:本研究得到教育部人文社会科学研究规划基金项目“虚拟学术社区科研数据流转演化机制及‘共享—重用’提升策略研究”(编号:23YJA870012)、江苏省社会科学基金项目“基于高价值专利的江苏未来产业前沿交叉技术识别与攻关机制研究”(编号:24TQB005)、江苏高校哲学社会科学研究重大项目“江苏省未来产业高价值专利智能识别与培育机制研究”(编号:2024SJZD065)资助。
作者单位
杨怡 南京信息工程大学管理工程学院 
万晨硕 南京信息工程大学管理工程学院 
李瑾 南京信息工程大学管理工程学院 
胡泽文 南京信息工程大学管理工程学院 
摘要点击次数: 5
全文下载次数: 11
中文摘要:
      针对现有国产生成式人工智能(GenAI)产品用户满意度研究普遍依赖问卷调查,以及难以刻画属性层面非线性影响的问题,本研究旨在构建基于在线评论数据的用户满意度分析与优化决策框架,系统揭示不同产品属性对整体满意度的影响机制及其改进优先级。以豆包、DeepSeek、通义千问、Kimi和文心一言等主流国产GenAI产品为研究对象,首先通过LDA主题模型识别用户关注的核心属性,并对属性绩效进行量化;随后利用SHAP方法估计各属性对用户满意度的影响程度;最后结合IPA模型对属性优化优先级进行分类。结果表明,用户关注的GenAI产品属性主要包括输出质量、效率、图片识别、多场景支持、自主化、文本创作、学习辅助和语音交互。其中,输出质量为优势属性,多场景支持与学习辅助为重点改进方向,图片识别为潜在改进属性,效率、自主化、文本创作和语音交互属于继续维持属性。本研究为国产GenAI产品功能优化提供了数据驱动的决策依据。
英文摘要:
      Existing studies on user satisfaction of domestic generative AI (GenAI) products largely rely on survey-based methods and struggle to capture nonlinear effects at the attribute level. This study aims to develop a user satisfaction analysis and optimization framework based on online review data, systematically revealing the impact mechanisms of different product attributes on overall satisfaction and their improvement priorities. Taking mainstream domestic GenAI products—including Doubao, DeepSeek, Tongyi Qianwen, Kimi, and Wenxin Yiyan—as research objects, we first identify core attributes of user concern through the LDA topic model and quantify their performance. Then, the SHAP method is employed to estimate the influence of each attribute on user satisfaction. Finally, the IPA model is used to classify attributes according to their optimization priorities. Results show that users primarily focus on output quality, efficiency, image recognition, multi-scenario support, autonomy, text generation, learning assistance, and voice interaction. Among these, output quality is a strength attribute; multi-scenario support and learning assistance are key areas for improvement; image recognition is a potential improvement attribute; and efficiency, autonomy, text generation, and voice interaction are maintenance attributes. This study provides data-driven decision support for the functional optimization of domestic GenAI products.
查看全文   查看/发表评论  下载PDF阅读器
关闭

分享按钮