An Integrated Deep Learning Framework for NFT Image Classification and Customer Sentiment Analysis in Digital Commerce
Main Article Content
Abstract
As NFTs and digital transactions continue to soar in popularity, the demand for smart solutions that can process visual, digital assets and user input in the form of text has become increasingly apparent. Common machine-learning methods need to be preengineered in terms of features, and often fail to deal with very complicated visual patterns, domain-specific jargon, sarcasm, mixed emotions, and long-range context relationships. In this paper, an integrated deep learning approach to image classification of NFTs and customer sentiment analysis is presented. When processing, resizing, normalization and augmentation are performed prior to the extraction of high-level visual features, the image is used as the main input.If the image is used as the main input, resizing, normalization, and augmentation are performed beforehand, and then high-level visual features are extracted. For sentiment analysis in text, cleaning, tokenization, normalization, embedding, contextual modelling and multi-class prediction is used. The workflow processes both positive sentiment and negative sentiment as well as neutral sentiment, and uses accuracy, precision, recall, F1-score, confusion matrices, ROC curves and AUC as key evaluation metrics. The study also takes into account transfer learning, feature optimization, class imbalance, domain-specific vocabulary, interpretability, and deployment efficiency. The integrated approach shows improved accuracy and reliability of classification in the study, compared to the isolated and conventional approaches, and it offers a flexible foundation for digital-asset evaluation, recommendation, fraud and anti-fraud control, as well as customer experience management and decision support.