Research On Massive And Sparse Spatial-temporal Data Analysis And Its Applications

With the continuous development of shared economy and the application of Internet of Things,there have been unprecedented spatial-temporal data,such as sharing cycling trails and crowd flow.The analysis of spatial-temporal data,especially from the massive spatial-temporal data to obtain meaningful information,at present has gained extensive research and application in academia and industry.On the one hand,spatial-temporal data analysis usually uses feature engineering and ma-chine learning.However,most spatial-temporal data,such as urban data,usually have sparseness and other issues.In addition,part of the application of learning ability for spatial-temporal data usually have timeliness requirements,but the actual process of learning usually requires expen-sive material and time costs.On the other hand,the structured expression and semantic mining of spatial-temporal data have the opposite advantages in constructing and enhancing the explanatory,accuracy and correlation data analysis of the analytical model.Combining the traditional machine learning framework and structured knowledge is one of the important ways to further enhance the analysis of spatial,temporal data,there is a large research future.This paper introduces a number of methods and applications for sparse,massive and inter-pretable spatial-temporal data analysis.The main contents of the research include five aspects:(1)Obtain spatial-temporal data from social media,physical sensors and other data sources,and classify urban areas into regions with urban data as an example,and use feature engineering and semantic mining to extract spatial-temporal related features,such as crowd view,region traffic and other features to help solve practical problems.(2)This paper proposes a method based on the multi-view transfer learning to solve the prob-lem of spatial-temporal data sparseness in spatial-temporal data analysis.(3)This paper proposes an efficient spatial-temporal data processing method based on extreme learning machine to solve the problem of timeliness in application.(4)This paper proposes a spatial-temporal data analysis framework based on semantic mining.It constructs structured data representation through ontology modeling,quantifies knowledge by using graph mining and feature fusion,and through the fusion of machine learning to obtain the interpretation of the results.(5)Some of the above methods of spatial-temporal data analysis are applied in urban comput-ing,namely:urban waterlogging analysis and urban placement analysis.