Aspect Based Sentiment Analysis – An Incremental Model Learning Approach Using LSTM-RNN

Analysing user opinions have become a necessity in order to understand customer satisfaction and requirements. Sentiment Analysis is the process where framework is capable of analysing and predicting sentiments of a sentence. Several researchers have attempted to perform sentimental analysis with various classification techniques especially machine learning techniques, however still issues faced with respect to analysis of data and decision-making process. Hence, this work attempts to improve the accuracy of classification process by implementing an incremental based novel framework. In this work, sentiments of restaurant reviews are processed and this will help in identifying satisfaction level of customers automatically. Artificial intelligence is used by implementing Recurrent Neural Network and Long Short-Term Memory. The hybrid LSTM-RNN is capable of analysing and predicting polarity of aspects with maximum possible accuracy. Various transformations and pre-processing techniques are applied for cleaning text-data as per required input to classifier. From reviews, lots of different aspects are obtained and trained to the LTSM-RNN classifier. When latest reviews are fed into the classifier, sentiments of each aspects are identified as either positive, negative or neutral aspects. The obtained accuracy is seen to be around 76% which is comparable with existing algorithms. Apart from good accuracy, major outcome is - multiple aspects are correctly extracted from multi-sentence lengthy reviews.