Knowledge Graph Construction And Application For Product Information Security Assessment

The attacks on network products are unpredictable,various forms of vulnerabilities have been generated,and the number of vulnerabilities has increased year by year.The public sentiment is used to evaluate the quality of product information security,and the security problems of products are discovered in a timely manner to the security of cyberspace and user information and property is vital.Aiming at the challenges of the large variety of network products on the market,the massive and fragmented safety and quality review information,and the lack of relevance between the information,a knowledge map study for product information security assessment was conducted,it can integrates the safety information of massive network products for efficient analysis and get useful safety information and the main tasks as follows:(1)An entity recognition method based on enhanced features is proposed.This method is improved on the basis of the BiLSTM-CRF network model.First,the artificial feature template is used to extract local features,and then the deep learning network BiLSTM is used to automatically learn the global features.Two results are combined to form enhanced features.Then input it into the conditional random field model,and obtain the final entity sequence labels.(2)A method of distant supervision for relation extraction based on semantic similarity was proposed.The external knowledge base is used to align with the text set,and the training samples are automatically annotated to construct a corpus without manual annotation,which can improve the efficiency of annotation.Aiming at the problem that the distant supervision method can produce wrong labels,a method for removing noise by semantic similarity is proposed to measure the similarity between the related words and dependent words in the knowledge base to remove the mislabeled samples.Input the denoised training corpus into a segmented convolutional neural network,automatically learns the semantic features of sentences and performs relationship extraction.(3)A product vulnerability assessment algorithm based on S-curve function was proposed.This algorithm combines the number of vulnerabilities and the score of the vulnerabilities to give a function curve,which is very close to the actual vulnerabilities,and can evaluate the quality of the product easily and quickly.The experimental results show that the enhanced feature-based entity recognition method has a higher accuracy rate on product vulnerability-related data sets than other models;the distant supervision for relationship extraction method based on semantic similarity can effectively remove noise data and improve the effect of the relationship extraction model;The product vulnerability assessment algorithm of the S-curve function can reflect the overall product risk simply,quickly and accurately.