智能座舱情感设计:一种多模数据驱动的使用−发言−心理联合分析方法

    Affective Design for Smart Cockpits: a Multimodality Data-Driven Method for the Joint Analysis of Usage-Comment-Psychology

    • 摘要: 围绕智能网联汽车 (intelligent connected vehicles, ICV) 智能座舱的需求分析及情感设计,探讨了如何通过优化车机系统功能配置以提升用户情感体验,通过深入挖掘用户在产品使用过程中产生的网络评论数据和行为日志数据,以及从用户评论文档中抽取心理发言需求特征,提出一种“使用−发言−心理”感性工学分析框架。该框架采用多模态数据,将语言模型与计量模型有机结合。通过增量预训练大语言模型,以95.22%准确率预测用户文本中提及的应用类型、用户发言意图、发言情感极性3类发言需求特征标签。将以上特征连同使用行为特征、TextMind心理语言学特征共同纳入结构方程模型,扩展需求工程的分析范畴,以深入解释用户对智能座舱的情感需求。对合作新能源车企的2 048名车主进行4个月行为追踪并分析,实验结果验证了本文方法在揭示用户应用需求和情感倾向方面的有效性。相较于线性回归模型, “使用−发言−心理”模型具有较好的拟合优度,可进一步揭示行为与心理语言学特征之间的隐含关系。本文利用用户大数据进行驱动,获得行为、观点及心理多层面上特征的自动提取及关系解释,为深入揭示用户复杂行为与设计需求间的机理提供了可行的解决方法。同时为制造企业理解客户需求、迭代产品的方向提供了有价值的参考。

       

      Abstract: Focusing on the requirement analysis and affective design of smart cockpits for intelligent connected vehicles (ICVs), a "usage-comment-psychology" Kansei engineering analysis framework is proposed to optimize the functional configuration of in-vehicle systems to enhance the users' affective experiences. The proposed framework is established by deeply mining the web comment data and behavioral log data generated by users to extract the characteristics of the psycholinguistic demand. Multimodal data are adopted in the proposed Kansei engineering analysis framework to combine large language models with econometric models. Through an incrementally pre-trained large language model, three types of comment demand feature labels, namely, the application type, the comment intention, and the sentiment polarity, are predicted from user comments with 95.22% accuracy. These features are incorporated with the psycholinguistic features derived from TextMind and the usage behavior features into a structural equation model to expand the analytical scope of requirements engineering, providing deeper insights int users’ affective requirements for smart cockpits. As an application example, the behavior of 2,048 ICV drivers of a partner new energy vehicle company are tracked for four months. Results verify the effectiveness of the method in revealing users' application requirements and sentiment tendencies. Compared with the linear regression model, the "usage-comment-psychology" model demonstrates superior goodness-of-fit and further reveals the implicit relationship between behavior and psycholinguistic features. Driven by the user big data, the method realizes automatic extraction and relationship interpretation of multiple dimensional features such as behavior, opinion and psychology, providing a feasible solution for revealing the mechanism between users' complex behavior and design requirements. It also provides a valuable reference for manufacturing enterprises to understand customer requirements and guide the iteration direction of their products.

       

    /

    返回文章
    返回