摘要
智能座舱的埋点数据蕴含丰富的驾乘人员行为信息,分析和识别其中具体的行为意图有利于深入洞察用户需求。考虑到现有依赖人工标注的行为意图识别方法存在成本高、主观性强及工作重复的问题,提出一种基于人工智能大模型的自动化标注和分类的方法。通过微调大模型Qwen2-14B,在多维度和多颗粒度上快速识别行为意图,提高了云端数据分析效率,为在车端实时响应用户需求奠定理论基础。
The event-tracking data of intelligent cockpit contains rich information about driver and passenger actions.Analyzing and identifying specific action intents can benefit deeper insights into user needs.Considering the high cost,strong subjectivity,and repetitiveness of current methods that rely on manual tagging for action intent recognition,a new method based on Artificial Intelligence model for automated tagging and classification is proposed.By fine-tuning the Qwen2-14B model,this approach could rapidly identify action intents across multiple dimensions and granularities,enhance the efficiency of cloud data analysis and lay a theoretical foundation for real-time response to user needs on the vehicle side.
作者
王文彬
尹佳伟
韩爽
刘航伶
何云廷
Wang Wenbin;Yin Jiawei;Han Shuang;Liu Hangling;He Yunting(Global R&D Center,China FAW Corporation Limited,Changchun 130013)
出处
《汽车文摘》
2025年第6期24-29,共6页
Automotive Digest
关键词
人工智能大模型
行为意图识别
智能座舱
埋点数据
自动化标注
Artificial Intelligence models
Action intent recognition
Intelligent cockpit
Tracking data
Automated tagging
