基于对比阈值的大数据流特征量最优挖掘算法

Optimal Mining Algorithm for Big Data Stream Feature Quantity Based on Comparison Threshold

针对因大数据流基数大、特征难提取导致的闭合模式下数据挖掘效率低、质量差的问题,提出一种基于知识图谱的大数据流闭合模式挖掘算法。结合闭合模式下大数据流的实体-关系-实体关系,通过迁移学习算法构建特征提取器,采集在该模式下不同特征数的输出标签,计算不同标签数据经过一次输出的特征损失量,以此作为损失函数的初始输入值,即为挖掘算法的初始参照。将大数据流在闭合模式下的标准元素作为挖掘目标,按照数据特征量大小建立状态序列,查找挖掘目标与序列中元素的相关性大小和对比阈值,结合目标点的映射关系得到最佳挖掘量,完成有效挖掘。实验结果表明,所提方法的挖掘精准度较高,耗用时间少,具备一定的实用价值。

At present,the efficiency and quality of data mining is low in closed mode.Therefore,based on knowl-edge maps,an algorithm for mining big data stream in closed mode was proposed.Combined with the entity-relation-ship-entity relationship of the big data stream in closed mode,a feature extractor was designed by the migration learn-ing algorithm.Then,the output tags with different feature numbers were collected in this mode.Meanwhile,the feature loss of different tag data after the output was calculated as the initial input value of the loss function,that is,the initial reference of the mining algorithm.Moreover,the standard element of the big data stream in closed mode was used as the mining target,and then a state sequence was constructed according to the size of the data feature.Finally,the cor-relation and contrast threshold between the mining target and elements in the sequence were calculated.Combined with the mapping relationship between target points,the best mining quantity was determined.Thus,we completed the effective mining.Experimental results show that the proposed method has higher mining accuracy and certain practical value.