Entity Relation Model In The Field Of Network Security Research And FPGA Implementation Scheme

With the rapid development of artificial intelligence in recent years,network security applications based on neural networks have also attracted widespread attention from academia and industry.Entity relation extraction can extract and mine valuable information from network security text data,and analyze its association relations.So as to improve the analysis ability of network security incidents.Most of the existing neural network models for entity extraction are mainly implemented by software.In order to improve real-time performance and meet the application of certain specific scenarios of network security,it is necessary to study the hardware implementation of network security applications based on neural networks.At present,most deep learning applications are based on general-purpose processor GPUs,but GPUs are not designed for machine learning,and cost,power consumption,and flexibility are relatively high.On the other hand,FPGA can be configured repeatedly and has a highly parallel structure suitable for neural network model realization.Its flexibility,stability,and security are high,and its energy consumption ratio is much better than that of GPU.Therefore,the realization of neural network model based on FPGA becomes the research direction of academia and industry.This article mainly focuses on the entity relation extraction model in the field of network security and its FPGA-based implementation scheme.Due to the division of labor and time relation of the project,it mainly focuses on the previous part of the research.The specific work is as follows:First,the type of entity relationship in the network security field is designed.We extensively refer to the structured threat intelligence expression STIX2.0 and the unified network security entity UCO2.0,and combine the actual network security text entity relationship categories to construct 8 types of entity relationship types in the network security field.Second,a Dense PCNN-ATT entity relationship extraction model oriented to the field of network security is proposed.This model mainly introduces a densely connected network to solve the problems of the current mainstream PCNN model,maximizes the feature reuse,and improves the effect of entity relationship extraction.In view of possible over-fitting problems,L2 regularization and Dropout are used to prevent over-fitting problems.The entity-relation extraction model is implemented in high-efficiency memory on GPU,and it is verified through comparative experiments that the model proposed in this paper can achieve high accuracy in general domains and network security data sets.Third,construct a data set in the field of network security.In view of the scarcity of data sets in the field of network security,this thesis uses crawlers to automatically crawl and collect secure text corpus,and perform operations such as cleaning and filtering.Then,the text data is automatically annotated through remote supervision and learning,and the initial annotation data set is generated.Check the system,filter the incorrectly marked examples,and get the final data set.Approximately 5,000 annotated corpus training data were constructed.Fourth,based on FPGA,the convolution and pooling module,activation function,model implementation overall framework and experimental scheme of the simple network security field entity relation extraction model are designed.