Optimized Deep Learning Framework for Opinion Mining of Customer Feedback Using Enhanced CNN Model

Abstract

Analyzing sentiment is one of the most crucial jobs for comprehending customer feedback and can help the decision-making process in different domains, including the restaurant domain. Using standard CNN, ResNet-50, RNN, and LSTM models as existing approaches may result in limited accuracy, poor generalization, and reduced computational efficiency. To overcome these limitations, we propose the Optimized Sentiment Analysis Using the Enhanced CNN (OSA-UEC) framework to efficiently and accurately classify customer reviews as neutral feelings, negative feelings, and good feelings. One uses pre-trained GloVe embeddings to extract features after performing advanced pre-processing on the text data. The Enhanced CNN model consists of dropout layers, max-pooling processes, and numerous convolutional and dense layers to capture local and global textual features efficiently. The flowchart diagram in Fig. 2 illustrates that the OSA-UEC algorithm coordinates training, testing, and performance metrics. The Zomato dataset's experimental findings demonstrate that the suggested approach produces superior outcomes with a state-of-the-art accuracy of 98.73% compared to other existing methods. A scalable and robust solution is usable for real-world scenarios like tracking customer sentiment and service quality to give best-suited recommendations. Extending to multi-domain, multi-lingual datasets and optimizing this framework for real-time deployment on resource-constrained platforms are promising future work avenues to improve its practical utility.