中文网络文本的地理实体语义关系标注与评价

Web texts contain mass of geo-semantic information, which provide opportunity for geographic knowledge services. Besides, the semantic relation corpus of geo-entities is an indispensable resource for geo-semantic understanding. For the web texts with a lot of noises, building a semantic relation corpus of geo-entities effectively and efficiently becomes a big challenging research question for geographic information science. By combining geographic information science and computational linguistics, this project focuses on building and evaluating a semantic relation corpus of geo-entities with Chinese web texts. In particular, the research content is composed of three parts, including the annotation schema, intelligent annotation algorithm and automatic assessment method of the semantic relation corpus of geo-entities. This project aims to theoretically and methodologically improve building semantic relation corpus of geo-entities with Chinese web texts, and break the limitation of traditional GIS on geo-semantic understanding for web texts with noises, as well as achieve the original innovation of corpus automatic assessing, so as to promote the capability of geographic knowledge services.

网络文本蕴含大量的地理语义信息，为地理知识服务提供了巨大潜能。地理实体语义关系语料库是实现地理语义理解的先决条件。针对多噪声的网络文本，快速构建高质量的地理实体语义关系语料库，对地理信息科学的理论与方法提出了严峻挑战。本项目以中文网络文本的地理实体语义关系语料库构建与评价为主题，通过地理信息科学与计算语言学的学科交叉，开展地理实体语义关系的标注体系、智能化构建方法、自动质量评价的研究，完善中文网络文本的地理实体语义关系语料库构建的理论方法体系，突破传统GIS理论与方法无法快速处理网络噪声文本蕴含地理语义理解的局限，实现定量与定性相结合的语料库自动评价方法的原始创新，增强地理信息语义理解能力。

地理实体语义关系是地理信息的重要组成部分，是支撑地理知识服务的关键数据，可形成以位置为关联的地理语义网络，帮助地理知识的传播、交流与共享。本项目围绕“地理实体语义关系的标注体系”“地理实体语义的评价体系”“标注语料库的构建方法”“标注语料库的质量评价方法”等关键基础问题开展了体系化研究，设计了经典语义关系与空间关系兼顾的标注准则，提出了实体语义标注的系列优化方法，为语料匮乏情况下的监督机器学习模型训练提供了一种有效的解决思路。按照既定的研究计划，本项目圆满完成了既定的研究内容，并进行了扩展应用。在项目资助下，发表学术论文9篇，申请国家发明专利1项，支撑研发平台2个，培养研究生3名，参加国际学术会议6人次。