Sentiment analysis (SA) is the process of evaluating human emotions, opinions and reviews expressed in text to detect the writer’s mental outlook towards a particular event, topic, product, service, etc. and assign a relevant sentiment. In SA, to highlight the correct words which contribute towards sentiment cognition is very difficult. Simulating this task of shortlisting of words by human observers is challenging due to complexity of human mind’s processing. The use of fuzzy entropy is proposed in our work as an innovative step to tap sentiment quotients of online movie reviews. We have proposed a novel approach of shortlisting of words that help in sentiment cognition using a combination of fuzzy entropy, k-means clustering and sentiment lexicon SentiWordNet. We have addressed this challenging task of simulating the human cognition of words by developing a model that recognizes sentiment based on fuzzy scores derived from SentiWordNet in an automatic manner. Experiments on two benchmark movie review datasets-IMDB and polarity dataset by Pang and Lee, with training by LSTM neural networks, yields high accuracy for our approach as compared to other state-of-the-art-methods of SA.