6 papers were presented at this group meeting.
| Paper and research focus | Presenter | Publication | Resources |
|---|---|---|---|
Large Language Models as Urban Residents: An LLM Agent Framework for Personal Mobility Generation本研究提出了 LLMob 框架,这是首个利用 LLM Agent(智能体) 基于真实世界数据进行个人移动轨迹生成的框架。通过“自我一致性(Self-Consistency)”的模式识别“检索增强(RAG)”的动机生成,解决了传统模型缺乏语义理解和难以适应突发场景(如疫情)的问题。 | arXiv 2024-10 | Not available | |
Learning universal human mobility patterns with a foundation model for cross-domain data fusion本文提出一种融合多源跨域数据与大语言模型的人类出行基础模型框架,用于构建可迁移、隐私保护且能高精度复现真实交通与出行模式的通用人类移动建模与仿真方法。 | Transportation Research Part C 2025-08 | Not available | |
DangerMaps: Personalized Safety Advice for Travel in Urban Environments using a Retrieval-Augmented Language Model本研究针对现有移动生成模型缺乏可控性的问题,提出了Geo-Llama框架。该框架利用大语言模型(LLM)并通过一种新颖的“访问打乱(Visit-wise Permutation)”策略进行微调,不仅能生成逼真的轨迹,还能强制执行特定的时空访问约束(如必须在特定时间到达特定地点),同时保持轨迹的上下文连贯性。 | arXiv 2025-10 | Not available | |
Beyond words: evaluating large language models in transportation planning | Geo-spatial Information Science 2025-04 | Not available | |
Leveraging large language models for citizen-centric urban_x000b_accessibility analysis: a case study using Airbnb reviews in DublinToday, we will focus on one of the most popular open-source LLMs: Flan-T5. The FLAN-T5 LLM was evaluated on a manually curated synthetic test dataset, achieving 84% and 86% accuracy in extracting public-transport and city-center travel times, respectively. | Spatial Information Research 2025-08 | Not available | |
GTA: Generative Traffic Agents for Simulating Realistic Mobility Behavior本文提出了一种名为 GTA 的生成式交通代理模型,利用大语言模型(LLM)驱动的基于角色的智能体,从人口统计数据生成虚拟人群,模拟其活动安排和出行方式选择。该模型在柏林规模的实验中得到验证,能够在一定程度上复现真实世界的交通模式,并为城市规划和交通政策评估提供了新的工具。 | arXiv 2026-01 | Not available |