Towards Text Mining in Climate Science:Extraction of Quantitative Variables and their Relations

Abstract

This paper addresses text mining in the cross-disciplinary fields of climate science, marine science and environmental science. It is motivated by the desire for literature-based knowledge discovery from scientific publications. The particular goal is to automatically extract relations between quantitative variables from raw text. This results in rules of the form “If variable X increases, than variable Y decreases”. As a first step in this direction, an annotation scheme is proposed to capture the events of interest – those of change, cause, correlation and feedback – and the entities involved in them, quantitative variables. Its purpose is to serve as an intermediary step in the process of rule extraction. It is shown that the desired rules can indeed be automatically extracted from annotated text. A number of open challenges are discussed, including automatic annotation, normalisation of variables, reasoning with rules in combination with domain knowledge and the need for meta-knowledge regarding context of use.