A hybrid SVD-HSV visual sentiment analysis system

Abstract

Image is worth a thousand of words. The use of images to express views, opinions, feelings, emotions and sentiments has increased hugely on social media. A lot of researches have been done for sentiment analysis of textual data. However, there is a limited work regarding visual sentiment analysis. In this paper, we propose a hybrid image sentiment prediction system, which combines low-level features and mid-level features of an image to predict the sentiment in different datasets. The results of the proposed hybrid system are better than using low-level or mid-level features individually. The results show that, the accuracy of the hybrid system exceeds the accuracy of using SVD only by 10% when being applied on photographic based images as in the KDEF dataset. Additionally, the accuracy of the proposed system exceeds the accuracy of using only HSV by 9% when being applied on social media images as in our collected and proposed dataset (SMI dataset). Another contribution of this paper is to avail the benchmarked dataset online for researchers.