Expressive Facial Characters Based on Indonesian Compound Sentence Using Multinomial Naïve Bayes Classifier

Abstract

Facial expressions are necessary for effective communication. In the field of animated facial films and game production, expressive facial characters are needed so that dialogue scenes can take place naturally as a human. To develop expressive facial characters like humans, these characters must be able to recognize emotions. Emotions from a text can be identified by classifying the text. This research aims to build an expressive facial character animation with the process of mapping dominant emotion classes from Indonesian sentences using the Multinomial Naive Bayes (MNB) model. A compound sentence is a sentence that has two or more clauses. The relationship between clauses is indicated by the presence of conjunctions. The classification process can produce complex emotional class probabilities (not just one emotion class). Emotion classes resulting from the classification process have different probability values. Therefore, a dominant threshold equation is needed to determine the dominant emotion classes. The dominant emotion classes of a compound sentence can consist of one emotion class. The combination of dominant emotion classes is called compound expression. In the development of 3D facial animation, the visualization of compound facial expressions is determined by the value of the associated Action Units (AUs). The results of this research indicate that MNB model can be used to map emotion classes based on the Indonesian compound sentence and the dominant emotion class can be determined using the dominant threshold equation. The dominant emotion classes as the basis for the formation of compound facial expressions.