Research on Transformer‐Based Triple Extraction Method for IoT Threat Intelligence

In response to the problems of heterogeneous text, complex entity relationships, and difficulty in extracting overlapping triplets of IoT threat intelligence, this paper proposes a Transformer based method for extracting triplets of IoT threat intelligence relationships and constructs a secure dataset for smart home scenarios. This method uses Transformer encoder as the backbone, integrates BiLSTM global interaction module to achieve global text correlation feature extraction, and adopts table filling strategy to complete triple joint extraction. The experiment was validated on WebNLG*, NYT public datasets, and self‐built IoT security triplet datasets. The results showed that the proposed method achieved the highest F1 value of 93.13%, and its robustness was significantly better than mainstream models such as Casrel, TPLinker, and CasDE in complex scenarios such as entity overlap and multiple triplets. This method can effectively improve the accuracy of structured extraction of threat intelligence, providing key technical support for the dynamic construction of IoT security knowledge graph and intelligent defense.