Sentiment Recognition by Rule Extraction from Support Vector Machines

Abstract

Affective computation allows machines to express and recognize emotions, a core component of computer games. A natural way to express emotion is language, through text and speech; computational methods that accurately recognize emotion in text and speech are therefore important. Machine learning techniques such as support vector machines (SVMs) have been used successfully for topic detection in documents and speech as well as for the identification of authors/speakers. SVMs have also been used for emotion detection in written and spoken communication, although with mixed success. An impediment to emotion extraction by use of support vector machines is that, after learning, it is not quite clear what has been learned. For instance, a gamer may acoustically respond to a character with fear and the SVMs that observe user behaviour confuse the sentiment (fear) with the character (e.g. an in-game persona). Successful emotion identification by support vector machines requires methods that ensure the recognition of sentiments without any confusion with certain topics or characters. This paper provides an introduction to affective computation and rule extraction from support vector machines, a set of techniques used for emotion recognition in text.