Research On Data Pattern Discovery And Object Evaluation Based On Complex Network

A new era of big data has huge amounts of data which contain potential,valuable information and instructions waiting for being studied and developed by data analysis methods to make efficient utilization.There are many modes can storage data,among which relational data is one of the most common storage mode.This paper takes relational data for research objects and by using complex networks methods to analysis data.We mainly focus on the study of data networking,key nodes discovery and data object evaluation.The main work we make are in as followings:(1)For networking the relational data,we propose a network construction method based on data features,which can effectively identify the key features from data.In this feature network,data features are abstracted as nodes and the probability of this data object on two features belongs which class is abstracted as link between two nodes.The feature network we construct can effectively guide us to discover different data patterns contained in massive data.Furthermore,in order to find the most significant features of classification results from the analysis of data objects,we use node importance ranking in complex network to find key features,and then we make Logistic Regression to key nodes.The results of classification experiments based on medical data show that because of the network structure is more robust than the data characteristic numerical value,so data feature network method can discover key features which are more significant for data classification.(2)For evaluating the data objects,we propose a data evaluation method based on fitness model of complex network.It is always worth continuing in-depth study of the problems like how to evaluate data objects objectively and effectively.Our method draws on the fitness model in complex networks to model the evolution of data objects.In addition,based on actual observation we introduce a network node deletion mechanism,and establish a method of evaluating data objects through network fitness.Experimental results based on school degree data show that: The fitness distribution of liberal arts,science and engineering teachers showed significantly different patterns,and these patterns also reveal some interesting phenomena of human resource.Above all,our method is expected to provide a new reference for data object evaluation.