Deep Attention Learning Mechanisms for Social Media Sentiment Image Revelation

Abstract

Sentiment analysis systems can handle social media images by interpreting the embedded emotional responses in those images. This represents an interesting and challenging problem that tries to figure out the high-level content of large-scale visual data based on algorithms devised from computer vision. This paper presents a system to analyze social media images and visualize the implied emotions from each image as (Happy, Sad, and Neutral). The objective of this work is to introduce a system model with features extraction basis utilizing some adequate technique of machine learning. The applied methodology is pivoted on implementing the required system through several steps of processing. This involves social media image displaying and video frames grabbing, image features extraction, then embedded emotions patterns classification and recognition utilizing a proper convolutional neural network (CNN). Flickr and Twitter datasets were utilized while the pertinent algorithm was developed using “Matlab2017b” platform. This can help social media users visualizing their interests besides forming a better scope of visualization. It will further assist companies in envisaging the mood of users/costumers towards their stock prices in order to set competitive prices for both sides. We design a Deep Attention Network Mechanisms (DANM) to achieve a higher level of social media sentiment image analysis and classify them as (Highly positive mood and highly negative mood). The DANM produces features maps basis utilizing the adequate focusing technique of machine learning based on a proper convolutional neural network (CNN). The proposed CNN training system has proven better results with respect to accuracy and efficiency in comparison with some other similar works. When experimentations on both real and synthetic datasets were conducted, the system showed a percentile improvement of about 14.2%. This system is applicable to a broad horizon of applications such as studying the emotional response of humans on visual stimuli, visual sentiment analysis algorithms and modeling, building machine learning-based robust visual sentiment classifier, as well as in most online websites that involve visual data mining for business intelligence, e-commerce, stock market prediction, political vote forecasts, and video gaming.