Using TextCNN Method to Extract Sentiment Tendency and Material Perception Mapping in Leather Product Consumer Reviews

Abstract

Accurately identifying the relationship between consumer sentiment and material perception is essential for intelligent product evaluation and data-driven manufacturing optimization. In engineering scenarios involving electromagnetic sensing and smart inspection systems, reliable semantic understanding of material-related information can provide complementary support for multimodal perception and decision-making. To address the problems of mixed sentiment expressions, long-range semantic dependencies, and ambiguous sentiment-material associations in leather product reviews, this study proposes an improved TextCNN framework integrating gated attention and multi-task joint learning. The proposed method dynamically filters salient semantic features through a gating mechanism and employs global attention weighting to reconstruct long-distance contextual relationships while simultaneously optimizing sentiment classification and material perception mapping using shared semantic representations. Experimental results demonstrate that the proposed model achieves a Macro-F1 score of 89.4% for sentiment recognition, an average semantic association accuracy of 79.9%, and a mapping coverage of 81.6%, outperforming representative baseline methods. The framework effectively enhances the interpretability of sentiment-material coupling and provides a practical semantic analysis strategy for intelligent material evaluation, multimodal information fusion, and engineering applications associated with electromagnetic sensing and advanced manufacturing systems.