Cross-discourse Development of Supervised Sentiment Analysis in the Clinical Domain

Abstract

Current approaches to sentiment analysis assume that the sole discourse function of sentiment-bearing texts is expressivity. However, the persuasive discourse function also utilises expressive language. In this work, we present the results of training supervised classifiers on a new corpus of clinical texts that contain documents with an expressive discourse function, and we test the learned models on a subset of the same corpus containing persuasive texts. The results of this indicate that despite the difference in discourse function, the learned models perform favourably. 1