Sentiment analysis is a categorization task that aims to identify polarity scores of comments shared by social media users, these comments can take either the form of image (like an emoji) or speech (like an audio) or text. Because of fast-growing in the use of social media like Facebook, Twitter, Instagram etc. the need for opinion extraction has increased at the same time. Opinion mining is used to help organizations and even individuals measure the public opinion (positive, negative, or neutral) of people about a particular event, topic, or product. In the past years, several approaches have been designed to supply the user with a performed categorization of emotions. These approaches have developed out from dictionary based approaches to machine learning and more recently to deep-learning. But these techniques still face two issues: the first issue is that the conventional deep learning algorithms struggle to proficiently detect and extract long-range dependencies in processed data and the second issue is the ingrained haziness and vagueness in natural language. Our principle objective is to resolve these both issues by integrating the fuzzy logic to handle the ambiguity inherent in natural language and the transformer DistilBert to effectively detect and extract the long-range dependencies in order to offer more effective opinion extraction to the user. Experimental findings obtained in terms of accuracy, F1, recall and precision prove that our approach handles accurately both issues in comparison with the state-art-methods.