Stream Tensor-based Multiple Clusterings Over Sliding Windows

In the era of big data,the Internet of Things and 5G,with the popularization and development of various types of smart terminals,smart applications and sensors,data is showing a trend of rapid and dynamic growth,and there are more and more applications that need to face streaming data.Although there is huge potential value in streaming data,because of its rapid growth,continuous and time-sensitive characteristics,how to efficiently and dynamically analyze streaming data has become an important issue.The tensor-based multiple clustering(TMC)method,as a more advanced method in the field of clustering in high-order data,can perform multi-modal analysis and mining of highorder big data in multiple different dimensions.At present,the research on TMC,and most of the clustering method theory is limited to static,small-scale data processing,and there is no relevant research on TMC for streaming data.By studying the two stages of TMC method: weighted tensor distance calculation and the final clustering,the sliding window mechanism is utilized to process stream data,and two challenges of weight tensor stream updating and stream clustering in TMC method are solved.Moreover,the trade-offs between accuracy and cost in weighted tensor streaming updating and streaming clustering are discussed,as well as the application scenarios and algorithm schemes respectively,and finally the stream TMC methods in different scenarios are present.First,through the study and derivation of the weight tensor learning method in TMC,two different methods for stream weight updating on higher-order tensors are proposed—the iterative method and the differential method.The two methods can both accurately and efficiently implement streaming incremental weight learning.The difference between the two methods is that the iterative method is better on accuracy,and the differential method is better on efficiency.On the other hand,the streaming update of the final clustering stage in TMC is deeply discussed and analyzed.According to the different focuses of the accuracy and efficiency,two methods seperately based on density peak clustering and K-medoids were proposed where the stream density peak clustering focuses on accuracy,and the stream Kmedoids focuses on efficiency.The four method modules can be combined into two stream schemes.The first combination of the differential method and the stream K-medoids method can meet the more efficiency-sensitive scenes,such as edge calculation and fog calculation.And the second combination of the iterative method and the stream density peak clustering method can meet the scenes where the accuracy is more sensitive,such as cloud computing.Finally,by simulating stream data with a large-scale data set,real experiments on the stream TMC method are conducted,and comprehensive evaluations of them are carried out using various evaluation criteria.The effectiveness of the proposed method is proved by the experimental results.