Social emotion classification aims to predict the aggregation of emotional responses shared by different users. Moreover, online comments are typically characterized by a sparse feature space, which makes the corresponding emotion classification task very difficult. Recently Hybrid Neural Network (HNN) has been proposed for Social Emotion Classification. However the still the reduction of sparse feature space from the emotion dataset becomes very difficult task. To solve this issue, major contribution of the work is to propose a new Quantum Behavior Particle Swarm Optimization based Sparse Encoding (QBPSO-SEn) technique to select optimal features from dataset. Properly, integrate a sparse regularization term to correct the expected activations of hidden units in Convolutional Neural Network (CNN). In order to handle sparse data problem presented by training model, the QBPSO-SC system is developed and it is used for improving the constancy of the CNN model. However, the regularization term with QBPSO- SEn is only applied to the pre-training phase of CNN to control the possible noise generated by teaching models. Additionally Kullback Leibler Divergence (KLD) based stochastic gradient descent method for transferring high-level semantic features from unsupervised learning models to CNN to enhance their emotion classification effectiveness. The experimentation outcomes prove that the presented CNN model achieved improved classification accuracy when compared to other methods such as Hybrid Neural Network (HNN) as well as Neural Network (NN) approaches.
Volume 11 | 06-Special Issue
Pages: 1636-1647