Anotace:
Emotions play big role in our everyday communication and contain important information. This work describes a novel method of automatic emotion recognition from textual data. The method is based on well-known data mining techniques, novel approach based on parallel run of SVM (Support Vector Machine) classifiers, text preprocessing and 3 optimization methods: sequential elimination of attributes, parameter optimization based on token groups, and method of extending train data sets during practical testing and production release final tuning. We outperformed current state of the art methods and the results were validated on bigger data sets (3346 manually labelled samples) which is less prone to overfitting when compared to related works. The accuracy achieved in this work is 86.89% for recognition of 5 emotional classes. The experiments were performed in the real world helpdesk environment, was processing Czech language but the proposed methodology is general and can be applied to many different languages.