Transforming sentiment analysis for e-commerce product reviews: Hybrid deep learning model with an innovative term weighting and feature selection

Improving user satisfaction by analyzing many user reviews found on e-commerce platforms is becoming increasingly significant in this modern world. However, accurately predicting sentiment polarities within these reviews remains challenging due to variable sequence lengths, textual orders, and complex logic within the content. This study introduces a new optimized Machine Learning (ML) algorithm named Enhanced Golden Jackal Optimizer-based Long Short-Term Memory (EGJO-LSTM) to perform Sentiment Analysis (SA) of e-commerce product reviews. This SA method comprises four critical stages: data collection, pre-processing, feature selection, feature extraction, and lastly, sentiment classification. The initial step involves utilizing a web scrapping tool to collate customer product reviews from various e-commerce websites. The collected data is subjected to a pre-processing phase to refine the scraped information. The pre-processed data then undergoes term weighting and feature selection processes by applying Log-term Frequency-based Modified Inverse Class Frequency (LF-MICF) and Improved Grey Wolf Optimizer (IGWO). In the final stage, the refined IGWO data is fed into the EGJO-LSTM model, which then classifies the sentiment of the shopper reviews into negative, positive, or neutral classes. Performance analysis was conducted using a prompt cloud dataset from Amazon.com, comparing the proposed classifier with state-of-the-art ML models. The metrics, such as precision, accuracy, recall and F1-score, were used to compare the performance. The results demonstrate that the EGJO-LSTM outperforms other models in sentiment classification. The proposed strategy is 25% and 32% better than the traditional and hybrid methods in terms of precision and accuracy. Further observations showed that when using the recommended LF-MICF weighting method, the EGJO-LSTM surpassed the performance of the state-of-the-art methods.