Microtext Annotation

Natural Language Processing

Karimi, Sarvnaz ; Yin, Jessie

CSIRO

DOI  :  10.4225/08/584af3a28ce23
RP-ID  :  EP13703

学科分类：地球科学(综合)

澳大利亚|英语

来源: CSIRO Research Publications Repository

【 摘 要 】

As microblogs have become a comprehensive repository of real-timeinformation, discovering whether or not a microtext, such as a tweet,contains useful information, and if it does what kind of information itconveys, is important for many social media mining applications. Ingeneral, text mining systems based on machine learning techniquesrequire training on examples of human-annotated text. Annotationidentifies information of interest as defined by the goal of anapplication.We are interested in annotating microblog posts (i.e. tweets) that arerelevant to specific events: natural disasters such as earthquakes orcyclones, and man-made disasters such as terrorist attacks or riots,that affect a large number of people or community. We developed anannotation scheme to cover three main aspects : what, when, and where. We report our annotation scheme, guidelines for annotators, the data used for annotation, and how this data was collected.