Research Of Textual Emotion Recognition Based On OCC Model

Text is a particularly important modality for sensing affect because the bulk of computer user interface today are textually based. Tranditional textual classification usually focuses on mapping text into the established topics like gym , economy, politics and etc. Nowadays more and more attention were paid to none topic textual analysis. Textual emotion classification which is also called textual emotion recognition means recognizing the underlining affective information from the text. It has became the key part of human machine interaction.Previous approaches to textual affect sensing have employed keyword spotting, lexical affinity, statistical methods, and hand-crafted models. For keyword spotting and lexical affinity, text is classified into affect categories based on matching affective words with the words in affective lexicon. Statistical text classifiers only work with acceptable accuracy when given a sufficiently large text input. Commonsense approach classifies the text into emotion categories using the large-scale real-world knowledge about the inherent affective nature of everyday situations. These approaches can to some extent effectively analyze the emotions of the text. However, because human emotion is complicated and changeable, recently the research mainly based on psychological theory, complete mathematic method did not emerge. There are still limitations on these approaches, for most of the models only focus on affective key words and they did not consider the personality factors.This paper presents a novel way for assessing the affective qualities of natural language by applying the cognitive theory of emotions known as OCC model. The paper defined a series of basic emotion reasoning rules and revised it to user's emotion reasoning rules on the basis of personality model. The basic emotion reasoning rules in the paper were built according to OCC model but simplified the 22 emotional rules of OCC model according to the specific feature of the text. To include the personal factor, the paper analyzed each dimension of the FFM (Five-Factor model) and modified the basic emotion rules based on the features of each dimension, and then built emotion reasoning rules for different people. An interactive method developed to reduce the misclassification rates. The accurate rates are improved using incremental learning which works on the use's feedback to dynamically adjust the knowledgebase and the threshod.To show the strength of the model, experiments were developed. The comprison shows that the chatting system with our model got higher accuracy than others, and that incorporating personality factors and using self-learning method give the model a further improvement.