Group Meeting | 2025-09-29
3 papers were presented at this group meeting.
Papers and research themes
| Paper and research focus | Presenter | Publication | Resources |
|---|---|---|---|
TransportationGames: Benchmarking Transportation Knowledge of (Multimodal) Large Language Models本研究提出了 TransportationGames，一个精心设计且全面的评估基准，用于评估交通领域的 (M)LLM。通过综合考虑实际场景中的应用，并参考布鲁姆分类法的前三个级别，我们测试了各种 (M)LLM 在记忆、理解和应用交通知识方面在选定任务中的表现。实验结果表明，尽管某些模型在某些任务中表现良好，但总体而言仍有很大改进空间。 |  @Yizhe Li | arXiv 2024-01 | Not available |
Uncovering the Social and Spatial Effects of Fare Cuts on Public Transport with Mobile Geolocation Data本研究利用覆盖全国的超大规模手机定位数据和时移式双重差分模型，评估德国 Deutschlandticket 票价补贴对出行量与出行距离的因果影响。结果揭示了政策整体提升出行活跃度，并对低房租及外籍人口集中地区带来更显著的出行促进作用，为公平高效的票价补贴政策提供实证支持。 |  @Qiang Xia | Transportation Research Part A 2025-08 | Not available |
Evaluating Urban Visual Attractiveness Perception Using Multimodal Large Language Model and Street View ImagesThis study introduces an automated framework for assessing urban landscape attractiveness by leveraging ChatGPT-4o’s advanced capability in image comprehension and linguistic reasoning to conduct automated, scalable and efficient aesthetic evaluations. The MLLMs approach improves evaluative efficiency while ensuring consistency in scoring criteria, thereby reducing subjectivity and offering a high-throughput, standardised solution for aesthetic analysis. ChatGPT-4o pairwise → Thurstone scale; validates vs 1.17 M Place Pulse votes (ρ = 0.76, 88 % faster than CNN). |  @IBRAHIM ABDI MOHAMMED | Buildings 2025-08 | Not available |