Design an image-based sentiment analysis system using a deep convolutional neural network and hyperparameter optimization

These days, sentiment analysis is a hot issue. The growth of sentiment analysis methods results from the exponential growth in data made possible by the emergence of social media networks. However, big data sentiment analysis with images is a challenging task. The most frequent obstacles with image-based sentiment analysis are noise, data scarcity, disappearing gradients, classification issues, and sentiment prediction. The system's performance is harmed, and sentiment prediction could produce poor outcomes due to distorted visuals and visual perception issues. This approach proposes the Deep Convolutional Neural Network with Hyper Parameter Optimization (DCNN-HPO) for correctly predicting sentiment analysis by optimizing the DCNN parameters. Moreover, image datasets are collected from the net source and trained in the system. All the images are pre-processed, and features are extracted using the VGG-16 network. Then, the extracted features are updated to the DCNN, and the weight parameters of the DCNN are optimized using the Krill Herd Optimization (KHO). Finally, perform sentiment analysis to classify positive, negative, and neutral sentiments from the input images. Thus, the designed model attained 98% accuracy, 99.12% sensitivity, and 0.2 s of execution time, which shows the efficiency of the developed model. Finally, the designed model accurately classifies the sentiment using input images.