Attention Based Swarm-optimized Quantum CNN for Sentiment Analysis in E-commerce Products

Abstract

In this study, a hybrid intelligent model based on deep learning techniques is presented, which exploits feature representation, optimization and enhanced feature learning capabilities. Quantum inspired Convolutional Neural Network, Efficient Dynamic Transformer, and Efficient Low-Rank Attention methods are incorporated into the design of the model to boost feature learning capabilities, context modeling capabilities and computational efficiency. First, the input data are cleaned and normalized to eliminate noise and normalize feature values. Quantum CNN is then utilized to generate feature representations, providing more efficient discriminability when compared with the traditional convolution approach. The extracted features are then used by the Efficient Dynamic Transformer to capture long-term contextual interactions, whereas Efficient Low-Rank Attention minimizes computation overheads through low-rank approximations. A series of experiments was carried out on benchmark datasets to evaluate the effective performance of the proposed method. The developed model yielded a mean classification accuracy rate of 99.47%, demonstrating superior performance to multiple state-of-the-art baseline models, including CNN, LSTM, and Transformer-based models. Other measures, such as precision, recall, F1 score and kappa score further validate the effectiveness of the proposed framework. Furthermore, experimental results show that the developed sentiment analysis model exhibits enhanced computational efficiency and generalization capabilities compared to current algorithms.