Multi-label Aspect Based Sentiment Analysis using Fuzzy LSTM

With increase in availability of social media side and cheap cost Internet facility, in present scenario people prefer to share their feelings and emotions in the form of comments or reviews on social media site. These comments are an important source for the reviewers and company people to know about their product from the customer point of view. The reviews mostly contain text words, emojis, number and special characters and thus, before processing the reviews per-processing technique is preferred to removed unwanted information. Many reviewers analyze the reviews into two classes i.e., positive and negative. In this paper, an attempt is made to analyze the reviews on multi scale i.e., in five scales namely great, good, average, bad, very bad. In this work, a combination of two machine learning techniques like LSTM (Long Short-Term Memory) and Fuzzy system is used to analyze the reviews. The LSTM is used to classify the reviews in binary class and based on the predicted polarity of the reviews, Fuzzy system helps in predicting the class in the scale of five. As labeled reviews are considered for analysis, the performance of the system is evaluated using confusion matrix and its associated parameters.