Emotions on Bengali Blog Texts: Role of Holder and Topic

The paper presents an approach to identify the emotions of the bloggers on different topics provided in the Bengali blog documents. The rule based identification of emotion holders and topics along with their corresponding emotional expressions forms the baseline system. A Support Vector Machine (SVM) based supervised framework is also employed to identify the three components from the blog sentences and it outperforms the baseline system. As the topic of a document is not always conveyed at the sentence level, the similarity between a document´s overall topic and sentential topic is measured through semantic clustering approach. Two different approaches are adopted to identify the many to many relationships among the holders and topics on Ekman´s six emotions. One is based from the perspectives of the holders and other is with respect to topics. The two way evaluation of Ekman´s six emotions achieves precision, recall and F-Score of 65.02%, 76.23% and 70.18% for 10 bloggers and 71.02%, 78.47% and 74.55% for 8 different topics on 512 test sentences respectively.