Interpretable Emotion Analysis Based on Knowledge Graph and OCC Model

Sentiment analysis or opinion mining has been significant for information extraction from the text. At the same time, emotion psychology also proposed many appraisal theories for emotional evaluations and concrete predictions. While sentiment analysis focuses on identifying the polarity, appraisal theories of emotion can define different emotions and view emotions as process rather than states. In real life, the mechanism of emotional generations and interactions is complicated. Only plausible polarity can't provide enough explanations for the emotional mechanism. Hence an explainable model is in demand during emotion inference and dynamical analysis. In this paper, an analysis framework is constructed for interpreting casual association based on the emotional logic. Knowledge graph is introduced into the appraisal theories for inferring the emotions and predicting the action tendency. The emotion knowledge graph levels: concept level and case level. The concept level can be built manually as an abstract based on the appraisal model of Ortony, Clore & Collins (OCC model). The inference and predictions can be implemented at this level. The case level includes entities, objects, events and cognitive relations between them that extract from the text through the modular functions. The elements in the case level can be linked to the abstract types in the concept level for the emotional inference. We test this emotional analysis framework on several datasets from the appraisal theory and the text of drama works. The results demonstrate that our framework can make better inferences on emotions and good interpretability for human beings.