In today’s world, social media becomes very important for human beings. Twitter is one of them and used as a famous social media platform through which users can express their opinions on various events/matters/objects. These opinions in the form of messages are called as tweets. In this paper, an algorithm is used to find and classify tweets positive or negative with accuracy toward a specific subject. This proposed system is using the training data set dictionary to observe the semantic orientation of tweets. The sentiment analysis in Twitter is used to know how people feel about an object at a particular moment in time and also tracks how this opinion changes over time. Sentiment analysis is most important part for many social media analytics tasks. This type of sentiment analysis is useful for consumers at the time of purchasing and finding the services of any product online as it is helpful to provide the opinion of others for the same product or service. It is also helpful for marketers and manufacturers to research public opinion for their organization/product and services. This paper presents a new concept of hybrid approach (Text and Image) for social media sentiment analysis. The hybrid approach consists of aggregating sentiments for both textual and visual contents. © 2018, Springer Nature Singapore Pte Ltd.