Item

Abstract

This thesis explores automatic sentiment analysis techniques with the overall goal of simulating human emotional perception of text. To approach this goal we collect an appropriately sized corpus of Grimm’s fairy tale texts, annotated for a rich set of emotional categories. We employ a full range of existing natural language processing tools for the German language on the collected texts in order to extract a large set of features. Finally, we apply two well-known machine learning algorithms (k-NN and Winnow) to perform various classification tasks on linguistic units as small as short phrases. The results we report are based not only on the pure automatic classification accuracies but also on human evaluation of resulting, machine generated, annotations. One distinctive feature of our approach is its maximal automatization. In order to bring in, analyze and automatically annotate a new text, no manual work is required. As such, we view this work as the first step towards the development of a more complex se
ntiment analysis system, which aims to simulate the actual human emotional perception of text.