Visual Sentiment-based on FER for Improving Feedback Analysis using Transfer Learning
Kriti Jhadi, Namita Tiwari, Meenu Chawla · 2024
Image sentiment analysis has recently become more popular than text sentiment analysis for making various judgments nowadays. Despite this, sentiment analysis based on images only provides sentiment polarities for input that might not accurately represent the user’s emotional state. The development of end-to-end image sentiment analysis methodologies has made extensive use of Transfer Learning(TL) techniques in recent times, and this trend is supposed to continue. Deep learning(DL) algorithms’ autonomous feature-learning capabilities have led to outstanding results in a variety of applications. Despite the challenges of image-based sentiment analysis, there is still significant potential for advancement. This research aims to improve feedback mechanisms using Facial Emotion Recognition(FER), by integrating emotional context with sentiment polarities,utilizing the well-known Deep Convolutional Neural Network(DCNN) VGG19, with its other deep features which enables the network to capture small fine details. By finetuning the VGG-19 architecture, Sentiment-Emotion detection and classification can be efficiently performed using the FER2013 dataset. The accuracy of the proposed method can reach up to $83.23 \%$, as demonstrated by a comparison of its results with various recent studies, This method makes it possible to analyze a person’s visual sentiments based on feelings and emotional patterns, which provides insightful information about the preferences, behaviors, and satisfaction levels of customers.