Sentiment Analysis for User Reviews Based on Improved Binarization Aquila Optimization with Self-Attention Bi-LSTM Model

Recently, people share their feelings and thoughts progressively on online review sites by customer feedback every day.However, classifying sentiments of customer opinions is challenging due to the lack of relevant information about specific sentiment class.To solve this problem, an Improved Binarization Aquila Optimization algorithm (IBAO) with Self Attention-Bidirectional Long Short-Term Memory (SA-BiLSTM) approach is proposed to enhance classification of user reviews effectively.Initially, the data from the Amazon product review dataset is preprocessed by employing four techniques: tokenization, stemming, lemmatization, and stop word removal.The preprocessed texts are then converted into numerical representation by Word2Vector method, while the features are extracted by the Term Frequency-Inverse Document Frequency (TF-IDF) technique.Then, feature selection is performed to remove irrelevant features by the Pearson Correlation Coefficient (PCC) and the proposed IBAO algorithm and classification of user reviews is performed by the SA-Bi-LSTM approach as Positive, Negative, and Neutral classes.The results of the proposed model for sentiment analysis are then evaluated based on the performance metrics with an Accuracy of 98.73%, which are greater than existing models namely, Recurrent Neural Network-LSTM (RNN-LSTM), and Enhanced Elman Spike Neural Network (EESNN).