Explicit and implicit oriented Aspect-Based Sentiment Analysis with optimal feature selection and deep learning for demonetization in India

Abstract

Aspect-Based Sentiment Analysis (ABSA) is a popular scheme that looks for the prediction of the sentiment of positive characteristics in text. The sentiment of text sequences is analyzed by deep neural networks and attained noteworthy results. Conversely, these models also have some problems with the limitation of past-training word embeddings and lack of communication between the context and the particular characteristic of the attention scheme. The main part of this task is to develop the novel ABSA concerning both explicit and implicit aspects using demonetization dataset reviews from India. Initially, the pre-processing of online tweets is performed by stop word removal, tokenization, lower case conversion, and stemming. Further, the explicit aspects are extracted, as it is simple to extract from the sentence and the polarity score is computed. A machine learning algorithm termed as Neural Network (NN) is utilized that helps for training the data regarding the implicit aspects, and further, helps to differentiate properly for the testing data with exact polarity score. Optimal feature selection is performed using the Self Adaptive Beetle Swarm Optimization (SA-BSO). These optimal features are given to a deep structured architecture called Recurrent Neural Network (RNN) with hidden neuron optimization by SA-BSO, which categorizes the demonetization reviews into positive, negative, or neutral. While taking the findings, the accuracy of the offered SA-BSO-RNN is secured at 4.67%, 6.56%, 3.54%, and 7.12% progressed than PSO-RNN, FF-RNN, CSA-RNN, and BSO-RNN, at 3-fold analysis for dataset 1. Results show that the designed ABSA concerning both explicit and implicit aspects using the demonetization method that provides enriched performance with diverse performance metrics.