<![CDATA[Sentiment analysis of customer reviews is vital for e-commerce, yet conventional CNNs and LSTMs face architectural constraints when processing unstructured Amazon Kindle Store feedback data. Specifically, CNNs prioritize local n-gram features over long-range semantic dependencies, while LSTMs often suffer from information dilution in the lengthy narratives typical of e-book product reviews. To address these identified research gaps and technical challenges, this study proposes an enhanced hybrid deep learning architecture integrating Gated Recurrent Units (GRU) with a Self-Attention mechanism. The methodology utilizes a large-scale dataset of 982,619 review instances, mapping five-point rating scales into binary sentiment categories while employing trainable GloVe embeddings and fixed-length sequences of 100 tokens to capture intricate domain-specific features. Furthermore, Random Oversampling is rigorously applied to mitigate inherent class imbalances between positive and negative reviews. Experimental results demonstrate that the GRU-Attention architecture achieves a superior classification accuracy of 0.973 and a training loss of 0.0930, significantly outperforming the CNN (0.961) and LSTM (0.948) baselines. The proposed model effectively prioritizes sentiment-critical tokens within reviews, attaining balanced Precision, Recall, and F1-scores of 0.973. These findings confirm the efficacy of attention-based recurrent networks in modeling unstructured textual data, offering stakeholders high-precision analytical insights to optimize customer satisfaction strategies and objectives.]]>
Copyrights © 2025
