Aspect-category based Sentiment Analysis on Dynamic Reviews

The emergence of social media has generated huge amounts of datawhich has led researchers to study the possibility of their exploration in order to identify their hidden knowledge. Two areas are being used for this: opinion mining and sentiment analysis. Sentiment analysis identifies and extracts subjective information in social material. In this paper we propose a method to process the user reviews and categorize them based on their aspects after which the aspect categories are recognized. Following this, the polarity of the reviews based on these categories are calculated, depending on the tone and adjectives used. In the proposed method, a Convolutional Neural Network (CNN) is trained and used for the task of aspect term extraction, which gives an f-measure of 88.97%. This CNN model uses general embeddings as well as domain specific embeddings for training purposes which are created using Word2Vec. Aspect term list and review pairs are processed to identify their aspect categories. Then the polarity of the review is detected using TextBlob. An aspect-category based summary is generated for the dynamic reviews input by the users using the deep neural network model.