论文部分内容阅读
传统方法在用户上下线高动态不定行为建模中没有考虑行为特征的遍历性和幂律特征，预测精度受限。提出一种基于先验概率特征检验和最小期望支持阈值的预测算法，采用n重伯努利试验得到上下线行为频繁数据集的幂律特征，构建高动态异常行为的转移概率平稳随机模型，采用马尔可夫链谱聚类排序方法对灰色预测算法的预测结果进行修正，实现对复杂网络节点频繁上下线高动态不定行为的准确预测。仿真结果表明，采用该方法进行频繁项集数据谱特征挖掘和复杂网络节点频繁上下线高动态不定行为预测，预测精度较高，执行效率提高，在大规模复杂网络管理和指导等领域具有较好的应用前景。“,”The traditional methods of high dynamic and uncertain behavior for online and offline behavior modeling do not consider the ergodic property and power law characteristic, the prediction accuracy is limited. A new prediction algorithm based on prior probabilistic characteristics test and minimum expected support is proposed, nth Bernoulli trial is taken for getting the power law feature, and the random transition probability model of high dynamic abnormal behavior is estab-lished. Markov chain spectral clustering ranking method is used for prediction correction, and the accurate prediction of the complex network nodes is obtained. The simulation results show that the frequent item sets spectral characteristics can be mined with this method, and the high dynamic and uncertain behavior can be predicted accurately, the execution efficiency is improved. It will have good prospects in large-scale complex network management and guidance.