A Joint Model of Entity Linking for RFC Protocols Knowledge Graph Construction

Abstract

Applying knowledge representation and reasoning to downstream tasks has been considered a promising research direction, as it enables semantic analysis of network protocols. Knowledge Graph is a new way of collecting knowledge, and building a protocol knowledge graph based on RFCs can help us study and analyze network protocols more effectively. However, automatically constructing a protocol knowledge graph from RFCs poses a major challenge, particularly in terms of extracting and linking protocol entities, due to the semi-structured nature of RFC documents. In this paper, we propose a model that combines a fine-tuned language model with an RFC Domain Model to link entities in RFCs to categories in the protocol knowledge base. Firstly, we design a protocol knowledge base as the schema for protocol entity linking. Secondly, we use heuristic methods to identify protocol entities and infer their descriptions from the nearby contexts of their header fields. Finally, we conduct comprehensive experiments on the RFC dataset using our joint model and baseline methods for protocol entity linking. Experimental results demonstrate that our model achieves state-of-the-art performance in entity linking on the RFC dataset, outperforming all baseline methods. In addition, we release a protocol knowledge graph, RFC-KG1.