Abstract: ［Purpose/Significance］Image-based sentiment analysis has gradually become a potential research hotspot. This paper intends to make a summary on the history and current situation of image sentiment analysis, which could be beneficial to promote relevant research.［Design/Methodology］This paper reviewed and offered comments on the technical methods of image sentiment analysis from two dimensions including the traditional image sentiment analysis method and the deep learning method.［Findings/Conclusion］It has been found that with the refinement of image sentiment analysis granularity, the further research will focus on the optimization of deep learning algorithm and annotation. At the same time, accelerating the opening process of sentiment labelled data-sets could help researchers to do further research in this field.［Originality/Value］This paper conducts an in-depth summary on the emphasis and development direction of image sentiment analysis, which could provide some reference for further research in this field.

Key words: Image sentiment analysis, Visual feature extraction, Image annotation, Deep learning