摘要
【背景】现有的出行推荐系统大多是根据特定目标,对各种交通方式的出行方案进行同质化推荐,未能考虑用户出行情景的异质性,导致难以满足用户个性化的出行需求。【目标】提出一种基于大语言模型的多模式出行方案个性化推荐算法,通过时空上下文感知,将多维异构的个性化出行情景表征为统一空间的嵌入向量,以更精准地刻画用户的出行需求,从而为用户推荐最匹配的出行方案。【数据】基于百度地图的历史出行大数据,提取用户出行查询记录、出行方案推荐记录、点击记录和用户属性等信息。【方法】利用BERT(Bidirectional Encoder Representations from Transformers)大语言模型将用户每次查询所对应的推荐出行方案集文本描述转化为语义向量,并与出行时段、用户属性等特征进行融合,构建综合表征用户个性化出行情景时空异质性的特征向量。将个性化出行情景特征向量输入多层感知器(Multilayer Perceptron,MLP)模型,预测用户最有可能选择的交通方式,作为推荐的最佳出行方案。【结果】基于BERT语义嵌入与MLP构建的多模式出行方案个性化推荐模型能够较为精确地辨识用户的差异化出行需求,其加权F1分数高于已有研究中使用相同数据集的其他模型。【应用】应用大语言模型感知用户个性化出行情景,可以显著提升出行推荐系统的智能化和个性化水平,对于优化出行服务质量,提高用户出行满意度具有重要作用。
[Background]Existing travel recommendation systems typically offer homogenized recommendations for various transportation modes based on specific objectives without considering the heterogeneity of users’travel contexts,making it difficult to meet personalized travel demands.[Objective]Propose a personalized multimodal travel recommendation algorithm based on a large language model.Through spatiotemporal context perception,the algorithm represents heterogeneous personalized travel scenarios as embedding vectors in a unified space,thereby accurately capturing users’travel demands and recommending the most suitable travel options.[Data]Based on historical travel data from Baidu Maps,user travel query records,travel plan recommendation records,click records,and user attributes were extracted.[Methods]The large language model BERT(Bidirectional Encoder Representations from Transformers)was utilized to convert the text descriptions of recommended travel plans corresponding to each user query into semantic vectors and integrate them with travel time periods,user attributes,and other features to construct a comprehensive feature vector that represents the spatiotemporal heterogeneity of individualized travel scenarios.The personalized travel scenario feature vectors were input into the multilayer perceptron(MLP)to predict the most likely mode of transportation that users will choose as the recommended optimal travel plan.[Results]The multimodal travel plan personalization recommendation model built with BERT semantic embedding and MLP can accurately identify the differentiated travel demands of users,with a weighted F1 score higher than that of other models using the same dataset in existing research.[Application]Applying large language models to perceive user-specific travel scenarios can significantly enhance the intelligence and personalization levels of travel recommendation systems,which is of great importance for optimizing travel service quality and improving user travel satisfaction.
作者
李文翔
丁龙远
张逸文
张行光
刘蓓
陈培焱
LI Wenxiang;DING Longyuan;ZHANG Yiwen;ZHANG Xingguang;LIU Bei;CHEN Peiyan(Business School,University of Shanghai for Science and Technology,Shanghai 200093,China;College of Computer Science and Technology,Zhejiang University,Hangzhou 310027,China)
出处
《交通运输工程与信息学报》
2025年第1期93-103,共11页
Journal of Transportation Engineering and Information
基金
国家自然科学基金项目(72471149)
教育部人文社会科学研究项目(24YJCZH147)
上海市哲学社会科学规划青年课题(2023ECK003)
上海市教育委员会“人工智能促进科研范式改革赋能学科跃升计划专项”。
关键词
智能交通
出行方案推荐
大语言模型
多模式出行
个性化出行情景
intelligent transportation
travel plan recommendation
large language mode
multimodal travel
personalized travel scenarios
