GeoGLUE

地理语义理解能力评测基准 GeoGLUE（GeoGraphic Language Understanding Evaluation）是由阿里巴巴达摩院自然语言处理组与高德联合发起提供的数据集，旨在推动地理相关文本处理技术和社区的发展。地理文本即描述地理实体、位置的自然文本，具有表达方式丰富、蕴含空间推理、知识强依赖等特点。此外，地理文本与现实地理世界的关联性也带来了诸多挑战。地理文本信息的自动化处理是许多地理语义应用的核心技术，本榜单提炼了其中多个典型场景：地图搜索、电商物流、政府登记、金融交通，并设计了六个核心任务：地理文本要素解析、地址地点切分、地址成分分析、地址实体对齐、POI召回、POI重排序。为了避免地理信息安全问题，涉及POI地址库部分我们基于全开源GIS系统OpenStreetMap标注，并人工编写了数十万Query。欢迎业界和学术界的同行们一起加入到GeoGLUE benchmark的建设中，一起来推动地理语义标准化数据集的发展。

The GeoGLUE (Geographic Language Understanding Evaluation) benchmark for geographic semantic understanding is a dataset jointly launched and provided by the Natural Language Processing Group of Alibaba DAMO Academy and Amap, aiming to advance the development of geographic-related text processing technologies and the relevant community.
Geographic text refers to natural language texts that describe geographic entities and locations, which feature diverse expression modes, implicit spatial reasoning, and strong knowledge dependence. Additionally, the correlation between geographic text and the real-world geographic landscape poses various challenges.
Automated processing of geographic text information is a core technology for numerous geographic semantic applications. This benchmark extracts several typical scenarios from these applications: map search, e-commerce logistics, government registration, and financial transportation, and designs six core tasks: geographic text element parsing, address and location segmentation, address component analysis, address entity alignment, POI recall, and POI re-ranking.
To mitigate geographic information security risks, the parts involving POI address databases are annotated based on the fully open-source GIS system OpenStreetMap, and hundreds of thousands of queries are manually curated.
We welcome colleagues from both industry and academia to participate in the development of the GeoGLUE benchmark and jointly promote the advancement of standardized datasets for geographic semantics.