Towards Affect-based User Models: a comparative study with various datasets, features and algorithms, for multi-labeled probabilistic affect detection

Authors

Abstract

In this paper, we use machine learning techniques to try to find the best possible combination of datasets, affect lexicons, features, and algorithms, for the automatic detection of affect in text. We consider as affective categories, the six basic emotions from Paul Ekman (anger, disgust, fear, happiness, sadness, and surprise). For the experiments, we count with three different datasets (news headlines, fairy tales, and blogs sentences), and two affect lexicons: WordNet-Affect and Roget's Thesaurus. From this collection of data, we compare the performance of two classification algorithms: Naive Bayes and Support Vector Machines (SVM). The results demonstrate that there are two combinations, in particular, that are more appropriate for the purpose under study. We evaluate and discuss the results from the perspective that in the future, we intend to build user affect models, based on the emotional information that can be collected from sentences, or texts, of the users.

Keywords

Machine Learning, Sentiment Analysis, Emotion Detection, Text Classification

Subject

Machine Learning

Conference

5th Symposium on Informatics (INForum 2013), Évora, Portugal 2013