EVGeoQA
收藏资源简介:
EVGeoQA是由北京师范大学与国防科技大学联合构建的动态多目标地理空间探索基准数据集,聚焦电动汽车充电场景下的实时空间规划问题。该数据集覆盖杭州、青岛和临沂三座中国城市,包含48,518条问答对,数据来源于国家电网充电站记录和高德地图POI信息,并通过融合人口热力图与路网数据的K-Means聚类算法生成用户坐标。数据集通过模板生成与LLM改写相结合的方式构建,严格验证充电站与周边兴趣点的语义匹配度。该数据集旨在评估大语言模型在动态地理空间环境中的多目标协同规划能力,推动空间智能在导航服务、城市计算等领域的发展。
EVGeoQA is a dynamic multi-objective geospatial exploration benchmark dataset jointly constructed by Beijing Normal University and National University of Defense Technology, focusing on real-time spatial planning problems in electric vehicle (EV) charging scenarios. This dataset covers three Chinese cities, namely Hangzhou, Qingdao and Linyi, and contains 48,518 question-answer pairs. The data is sourced from State Grid charging station records and Amap POI information, and user coordinates are generated via the K-Means clustering algorithm that integrates population heatmaps and road network data. The dataset is built through a combination of template generation and LLM rewriting, with strict verification of the semantic matching degree between charging stations and surrounding points of interest. It aims to evaluate the multi-objective collaborative planning capability of large language models (LLMs) in dynamic geospatial environments, and promote the development of spatial intelligence in fields such as navigation services and urban computing.

- 1EVGeoQA: Benchmarking LLMs on Dynamic, Multi-Objective Geo-Spatial Exploration北京师范大学·人工智能学院; 北京师范大学·教育应用人工智能北京市重点实验室; 北京师范大学·教育部智能技术与教育应用工程研究中心; 国防科技大学·计算机科学与技术学院 · 2026年



