Research On Cyber Security Capability Based Training Subject Recommendation

Improving cyber security capability is the main objective in cultivating students majored in cyber security,and personalized cyber security capability training plays an important role in implementing it.Current training subjects recommendation method,which is based on user preference,could not be used in our circumstance.Because it will occupy too much teaching attention when organizing such exercises,causing problems like high costs,low efficiency etc.Concentrated on problems existed in cyber security exercise,this paper proposed a thorough personalized training subjects recommendation plan which is based on quantitative evaluation of cyber security techniques.There are three steps in the plan,including evaluating cyber security technique quantitatively from training data,filtering training subjects set according to capability similarity,extracting sequence pattern from data above.This paper contributes as follows:1.Provide a step function based capability evaluation algorithm concerning the problem of lacking myopic correlation in training data existed in current algorithm.This paper firstly verifies the defects the problem existed in current capability evaluation algorithm,which results in huge error in evaluating capability and then incorporates myopic correlation information into capability evaluation algorithm through step function to rectify the problem in those algorithms.The experiment compares our algorithm with existed one and the result turns out that the average error ratio decrease by 11.73% and highest error ratio decrease by 16.25% when compared with known methods.2.Provide a double stage sequence similarity measure framework concerning the low accuracy problem existed in current algorithm.The existed sequence similarity measure algorithm could not meet the need in education recommendation because of high false positive.In order to deal with problem above,this paper comes up with a new method which includes two stage with the first of which filters redundant time series with Local Sensitivity Hashing technique,and the second of which measures the similarity accurately.The experiment result implies that the average precision increase by 16.34% in subsequence matching and highest precision increase by 20%.3.Provide a frequency distribution based sequence pattern extraction algorithm(F-prefixSpan algorithm)concerning the time consuming problem in current sequence pattern extraction algorithm.The current sequence pattern extraction algorithm is very time-consuming,making it hard to be used in a real-time education recommendation environment.This paper take advantage of the distribution information to prune the insignificant operations,improving the search efficiency.The experiment results turns out that our proposed algorithm could extract sequence pattern effectively,which save time about 6.2% when compared with current algorithms.A complete solution is proposed to implement training subjects recommendation in cyber security area.The design and workflow of recommendation system is depicted in this part to show the details,and SUS method is used to evaluate our propose scheme with the SUS score reaches 73.625,implying the effectiveness in improving cyber security capability.