摘要
为了准确判断软件的可信性,针对软件预期行为轨迹中的软件监控点,该文提出了一个基于分级属性的软件监控点可信行为模型。首先,依据软件监控点各属性在可信评价中的作用范围将属性分级,构建各级属性的可信行为模型。其次,针对场景级属性,对同一监控点的训练样本进行区分,提出了一个基于高斯核函数的场景级属性聚类算法;针对单类训练样本,提出了基于单类样本的场景级属性权重分配策略。最后,实验分析表明:基于分级属性的软件监控点可信行为模型能够准确地对监控点的可信性进行评价;对于场景级属性可信模型,采用基于高斯核函数的场景级属性聚类算法具有更低的分类错误率,基于单类样本的场景级属性权重分配策略具有更优的可信性评价效果。
In order to estimate the software trustworthiness accurately,a trustworthy behavior model for software monitoring point based on classification attributes is proposed for the software monitoring point in the expected behavior trace of software.Firstly,the attributes of software monitoring point are classified according to the sphere of action during the trustworthiness evaluation and the trustworthy behavior model of each attribute level is constructed.Secondly,for scene level attributes,a clustering algorithm of scene level attributes based on Gaussian kernel function is presented considering the distinction of training samples of one monitoring point,and a weight distribution strategy for scene level attributes based on one-class samples is proposed for one-class training samples.Finally,experiments and analyses show that： the model can evaluate software monitoring point accurately;For trustworthy behavior model of scene level attributes,the clustering algorithm has lower classification error rate than other clustering algorithms,and the weight distribution strategy has better effect of trustworthiness evaluation than other methods of weight distribution for one-class samples.
出处
《电子与信息学报》
EI
CSCD
北大核心
2012年第6期1445-1451,共7页
Journal of Electronics & Information Technology
基金
国家自然科学基金(60873203
61170254)
空天信息安全与可信计算教育部重点实验室开放基金(AISTC2009_03)
河北省杰出青年基金(F2010000317)
河北省自然科学基金(F2010000319
F2011201039)资助课题
关键词
软件行为
监控点
可信性评价
聚类
属性权重
Software behavior
Monitoring point
Trustworthiness evaluation
Clustering
Attribute weight
