Hierarchy Affective Knowledge Enhanced Emotion Recognition Model

Fine-grained emotion recognition models are better at capturing diverse emotions than traditional approaches. However, the huge number of category and interconnections of Fine-grained emotions pose challenges for traditional models. Existing fine-grained emotion recognition models overlook leveraging deep sentiment knowledge, such as the emotion hierarchy, in prediction models. To address this problem, this paper proposes a novel Emotion Recognition model based on Hierarchy Graph Convolution Networks (HGCN-EC). The HGCN-EC model comprises three modules: a semantic information module, an emotion hierarchy knowledge module, and a knowledge fusion module. The semantic information module extracts the semantic features from textual data; the emotion hierarchy knowledge module organizes fine-grained emotions into a tree-like hierarchy and calculates the transition probabilities between emotions using Bayesian statistical inference; the knowledge fusion module leverages a graph convolutional network to effectively integrate the hierarchy knowledge with the semantic information for emotion prediction. Comparative experiments on the GoEmotions dataset show that the proposed HGCN-EC outperforms baseline methods in the task of fine-grained emotion recognition.