Network_Protocol_Entity_Extraction_Based_on_Few_shot_Learning

Abstract

Knowledge Graph is a new way of knowledge collections, and building a protocol knowledge graph on RFCs can help us study and analyse network protocol better. Protocol entity extraction is one of the keys to constructing the network protocol knowledge graph.Because RFC (Request For Comments) contains detailed descriptions of basic Internet communication protocols, the protocol entities of the network can be obtained from it. However, the document format and wording are not uniform, which leads to the inability to complete the extraction task of network protocol entities based on traditional rule information extraction methods. Therefore, this paper proposes a network protocol entity extraction method based on Few-Shot Learning. This method can use a very small amount of labeled samples to extract network protocol entities from a large number of unlabeled samples and maintain high recognition accuracy. This method firstly mines as many potential network protocol entities as possible in the RFC document, and secondly performs accurate re-identification of the identified potential network protocol entities. Experiments show that using 5 manually annotated RFC documents to train our model, the accuracy of network protocol entity extraction reaches 88.4%. Compared with the existing methods, this method has higher accuracy and better robustness in terms of network protocol entity extraction, and it also has better identification ability for network protocol entities that have not appeared in the training set.