Social media has become a slogan in emotion and sentiment analysis. In today’s era Social media networking sites are almost used by everyone. Social media users share their feelings, thoughts, and experiences with other people by short messages. Short messages are composed of emoticons, slangs, noises, irrelevancies and words. Thus, preprocessing becomes the challenging task for Sentiment analysis. This experiment is performed to evaluate the impact of pre-processing on social data for sentiment classification particularly for slang words. This paper focused on identification of important slang words and to evaluate their impact on sentiment analysis of social media posts. The proposed scheme collects bigrams, trigrams of slang and exploits different features for better results of sentiment classification. N-grams are used for bindings and conditional random fields (CRF) to determine the importance of slang words. Experiments declare that this proposed scheme increases the accuracy of Sentiment analysis.