Cybersecurity entity recognition for blockchain of things via Hierarchical Attention Mechanism

Abstract

Abstract With the integration of blockchain technology and the Internet of Things (IoT), the blockchain of things (BCoT) has received more attention. Because of the lack of efficient security mechanisms, the number of security incidents aimed at BCoT has been growing exponentially. The traditional cybersecurity analysis methods can utilize cybersecurity knowledge graph to extract threat intelligence information with fine granularity for BCoT. Named entity recognition (NER) is the primary task for constructing cybersecurity knowledge graph for BCoT. Traditional NER models make it difficult to determine entities with complex structures and ambiguous meanings in BCoT. It also cannot efficiently extract non-local and non-sequential dependencies between the cybersecurity entities. So, the traditional NER cannot be directly applied in the field of BCoT. In this paper, we propose a novel Cybersecurity Entity Recognition model based on Hierarchical Attention Mechanism, denoted as CER-HAM, to extract cybersecurity entity in the field of BCoT. CER-HAM composes the self-attention mechanism with the graph attention mechanism to capture non-local and non-sequential dependencies between cybersecurity entities. Based on those dependencies, CER-HAM can accurately extract cybersecurity entity in the field of BCoT. In the end, the real cybersecurity dataset of BCoT is used to verify the efficiency of CER-HAM. The experimental results show that the F1-score reached by CER-HAM is better than the traditional entity recognition model.