Sentiment analysis (SA) is a mechanized strategy for finding and understanding the feelings depicted in text. Over the most recent decade, SA has altogether expanded in ubiquity in the Natural Language Processing (NLP) people group. Because of the inescapable utilization of social media and web stages, SA is presently urgent for firms to get client info and shape their promoting approach. The motivation behind this study is to utilize a deep learning-based way to deal with natural language process to give an exhaustive investigation of the opinion of sentiment of social media content. The technique then builds a comparable model that may be applied to replicate the relevant social process. In instance, by dynamically choosing the most significant phrase in the current state depending on the information presented, the network can effectively recognize the significant material that is always changing. This prompts the semantic comprehension of the total sentence toward the finish of each phase of the interaction. Besides, an absence of context oriented data will prompt an incorrect and muddled semantic portrayal of natural language on the grounds that relevant data is fundamental for that portrayal. In this paper, we take a gander at the unimodal and multimodal social network sentiment analysis algorithms and foster two models for text sentiment investigation and picture-text multimodal sentiment examination.