Deep Learning Approach for image sentiment classification using Convolutional Neural Network

Abstract

Sentiment analysis is a method for gauging online users' attitudes and beliefs from the words and images they share on platforms like Twitter and Instagram. Sentiment categorization is challenging since it requires identifying the emotions implicit in written or visual content. People's modes of expression of feelings vary with context and topic. In this research, we suggested a novel, deep-learning-powered method for feature extraction and selection. Deep CNN is a customized convolutional neural network used to evaluate the image features (luminosity, colour, histogram, autoencoder, etc.) that were taken from the original picture. Multiple activation functions and optimizers have been employed for feeding CNNs, and the number of deep CNN layers, feature extraction size, and activation function have all been experimented with. Our proposed module obtains higher accuracy for RESNET-1001 over RESNET-50 and RESNET-100. In comparative analysis proposed model archives higher accuracy than exiting deep learning frameworks.