地理语义理解能力评测基准

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, a dataset for evaluating geographic semantic understanding capabilities, was jointly launched and provided by the Natural Language Processing Group of Alibaba DAMO Academy and Amap, aiming to promote the development of geographic-related text processing technologies and the corresponding community. Geographic text refers to natural language that describes geographic entities and locations, which is characterized by diverse expression forms, requirements for spatial reasoning, and strong dependence on domain knowledge. Furthermore, the correlation between geographic text and the real-world geographic environment also brings various challenges. Automated processing of geographic text information is a core technology for numerous geographic semantic applications. This benchmark extracts several typical scenarios including map search, e-commerce logistics, government registration, and finance and transportation, and designs six core tasks: geographic text element parsing, address and location segmentation, address component analysis, address entity alignment, POI (Point of Interest) retrieval, and POI re-ranking. To avoid geographic information security issues, 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 created. We welcome peers from both industry and academia to join the construction of the GeoGLUE benchmark and jointly advance the development of standardized datasets for geographic semantics.

搜集汇总

背景与挑战

背景概述

GeoGLUE是由阿里达摩院与高德合作构建的地理文本理解评估基准，旨在推动地理文本处理技术的发展。该基准包含地址元素解析、地理实体对齐等六项核心任务，数据基于开源OpenStreetMap标注，以规避地理信息安全风险。