Sentiment analysis is an active research field as one of the most popular tasks of natural language processing, which aims to extract valuable information from various social platforms and extensive online texts to process and find people's attitudes in business and advertising, government, economic fields, and even political orientations. Hence, researchers have made many efforts in this field, which mainly refer to traditional approaches based on dictionaries, machine learning, and deep learning models. Graphs as a robust and interpretable data structure have been considered for applications of artificial intelligence models such as machine vision and natural language processing which are used for learning non-structured data like text or images. Although deep learning methods have achieved promising results in this field, due to problems such as assigning indecisive weights and high dimensions in feature extraction stages, they are still a “black box.” Meanwhile, graph neural networks (GNNs) are a particular type of deep neural network that are interpretable and flexible. Their adaptability in solving complex problems in data analysis with a graph structure has made them one of the most efficient methods in the last decade. Considering the large amount of textual information in social media and various online platforms, sentiment analysis or opinion mining aims to help marketing strategies for business owners and awareness of the attitude of public opinion in governments has become one of the crucial issues in today's modern societies. This comprehensive review focuses on GNN-based approaches in sentiment analysis and summarizes the recent state-of-the-art in this area. Also, we discussed their weaknesses and strengths, and challenges on specific datasets. Our goal is to show the development process and the potential of GNN-based approaches in different problems of sentiment analysis compared to previous methods and to help find more effective directions for researchers interested in this field.
