基于频繁密度分布模式的不确定数据流查询方法

Query processing on uncertain data stream based on frequency density distribution pattern

针对当前不确定数据流相似性查询问题中不确定对象建模不准确的问题,提出了一种面向不确定数据流的相似性查询方法 HB-UTS。利用非参数估计方法对不确定数据流中的对象建模,得到不确定对象的密度函数。通过谱聚类方法挖掘密度函数的频繁模式,将挖掘后的模式抽象为语义表示的不确定数据流序列。在相似性查询阶段,通过高阶Markov的状态转移矩阵模型构建不确定数据流的索引结构,它在记录不确定数据流存储地址的同时还记录序列元素的存储概率,可有效提高数据流的分步输入查询效率。本文进行了真实与仿真相结合的方法,通过在随机化处理后的真实数据集上的实验以及与其他相似性查询方法的比较,验证了HB-UTS在处理大规模不确定数据流时较好处理能力以及实施效果。

To solve the defect of inaccurate modeling for uncertain objects in processing similarity query of uncertain data streams,HB-UTS method was proposed. Non-parametric estimation is used to model the uncertain objects to obtain the density function. The frequency pattern of density function is mined by spectral clustering method and the mined object pattern is abstracted as an indefinite semantic data stream sequence. In the similarity query phase,an index structure of the uncertain data stream is constructed by the state transition matrix model of high-order Markov.It also records the storage probability of the sequence elements while recording the storage address of the uncertain data stream to improve the step-by-step input query efficiency of data stream. To analyze the effect of this method in practical problems,a method combining reality and simulation was adopted. By the experiments on the randomized real dataset and comparing with other similarity query methods,it was verified that HB-UTS is very effective in processing large-scale uncertain data stream.