In present day’s social networking technologies are increased because of different user’s communication with each others. There are different types of networks are available in present situations like face book, twitter and LinkedIn. These are the valuable resources for data mining applications because of prevalence presents of different user’s information present in outside environment. Sentiment analysis is the process that defines attitudes, views, emotions and opinions from text, database sources and tweets. Sentiment analysis involves to categorize data based on different opinions like positive and negative or neutral reference classes. In this paper, we analyze different machine learning approaches to define sentiment analysis on social networks. This paper describes comparative analysis of existing machine learning approaches to classify text and other reference classes to evaluate different metric representations. And also this paper describes different machine learning methodologies like Naïve Bayesian, Entropy max and support vector machine (SVM) research on social network data streams. And also discuss major innovations to evaluate different procedures and challenges of analysis of sentiment or opinion mining aspects in present social networks.