Sentiment-to-Sentiment translation is a special case for Style Transfer. Style Transfer is emphasised on generating the opposite polar style in terms of emotions or sentiment. This results in the transfer of style successfully but loses the semantic context of the sentence. This is caused due to inefficient amount of data having these relevant paired sentences with polar styles. This talk focuses on generating unpaired dataset which preserves the semantic context during a style change using cycled Reinforcement Learning approach on parallel data having emotionalization and neutralization modules.