Chinese Semantic Structure Mapping for Textile Material Performance Data: An Improved BiLSTM-CRF Approach to Cross-Modal Intelligent Retrieval

Efficient semantic understanding and cross-modal retrieval of textile material data have become increasingly important as intelligent manufacturing systems integrate heterogeneous information from technical documents, physical measurements, and multimodal databases. Similar challenges also arise in electromagnetic wave characterization and antenna measurement platforms, where reliable semantic mapping is essential for heterogeneous data organization and retrieval. To overcome the limitations of Chinese semantic parsing in textile material databases, this study proposes an improved BiLSTM-CRF framework incorporating dependency syntactic rules and domain knowledge constraints for semantic structure mapping. A multimodal hash encoder is introduced to project textual features and physical property parameters into a unified embedding space, while maximum mean discrepancy optimization enhances cross-modal consistency. Furthermore, an improved R-tree indexing mechanism with dynamic weight adjustment is designed to accelerate complex query processing and maintain retrieval stability under high-concurrency conditions. Experimental results show that compound term boundary recognition reaches approximately 91.8%, average cross-modal similarity increases to 0.87, and response latency remains around 185 ms with 500 concurrent users. The proposed framework demonstrates substantial improvements in semantic parsing accuracy, multimodal feature alignment, and retrieval efficiency, providing an effective solution for intelligent textile material management while offering methodological insights for semantic organization and heterogeneous information retrieval in data-intensive electromagnetic sensing and propagation-related applications.