Automatic Learning of Medical Text Annotation Rules - a Case Study on Endoscopies

摘要

We present a method for automatic learning of text annotation rules which relies on association rules mining, in particular on a modified version of Apriori algorithm. The method starts from a set of texts written in natural language, each of them having associated one manually established label and aims to mimic the way human experts have established those annotations. The rules learned are basically pairs of words combinations and labels. They are further stored as JAPE (Java Annotation Patterns Engine) rules and can serve to annotating new texts. No natural language processing tool is employed. The method has been applied on a set of descriptions of endoscopies in Romanian language, with promising results.