"Sentiment analysis is a widely used application of Natural Language Processing. m of sentiment analysis is to extract sentiment hidden in a piece of text. Research k is divided into two modules, sentence sentiment analysis and Twitter sentiment analysis which are part of sentence level sentiment analysis. The objective of sentiment analyzer is to extract polarity of opinions associated with a target subject on sentence level rather than classifying whole document as positive or negative. In order to obtain sentiment about a certain target subject in a document, it is beneficial to extract semantic relationship between sentiment term and target subject in each individual sentence. There are three classification categories positive, negative and neutral. The proposed sentiment analyzer is tested upon 1000 random plain sentences of different domains which were without any conjunction words and found that it correctly classified 887 sentences. However, in complex sentences accuracy reduced to 70-75% because in some sentences target subject and sentiment providing term was far away located in sentence. Second model is created for sentiment analysis of tweets. In this model opinion hidden in a tweet is classified into 2 categories as positive or negative after filtering the tweet by removing any links, Hashtags (#), emojis or User handles (@). Movie reviews opinion analysis is chosen as domain of thesis work and 1000 random tweets of different movies have been selected in English language only and found that the model were able to produce classification accuracy up to 85-90%. Sentence sentiment analysis model is generic and can be implemented on any kind of text in English language but twitter sentiment analysis is domain specific. First, Naïve Bayes classifier needs to be trained using the corpus of same domain in which sentiment analysis of tweets to be done. In future, the approach of window method in sentence sentiment analysis can be enhanced so that those sentences in which target subject and sentiment giving term are located far away from the predefined window can also be covered. In classification of tweets, analysis of emojis that has been omitted, can be very useful to achieve more accurate results. Extraction of extent of polarity is omitted in present work which can give advance information to business analysts to decide their
future policies."