RDLSH: Adaptive Entity Recognition and Relation Extraction for IoT Knowledge Graph

With the rapid development of the Internet of Things (IoT), security issues are becoming increasingly severe. Malicious attackers use IoT devices to carry out network attacks, resulting in data leakage. The use of knowledge graphs effectively prevents and resists attacks through deep mining and association analysis in security situation awareness and threat prediction. Entity recognition and relationship extraction are the core steps in the construction of knowledge graphs. They are used to automatically extract meaningful entities and relationships from massive data and perform reasoning, but they still face challenges in accuracy and computational cost in extracting long texts and complex relationships. To address these issues, this paper proposes the RDLSH model for processing of local context, low-frequency entity recognition, and global semantic associations. Based on the Reformer architecture, it dynamically adjusts the local sensitive hashing parameters, and combines the multi-head attention mechanism to achieve good performance in capturing cross-paragraph and long-distance dependencies, and efficiently handles entity recognition and relationship extraction tasks. In addition, the RDLSH model introduces reversible residual networks and bidirectional transfer mechanisms to optimize the memory usage of large-scale data processing and improve computational efficiency. Experimental results show that the RDLSH model not only improves the accuracy of entity and relationship extraction, but also enhances the cross-sentence dependency processing capability and computational efficiency.