| With the advent of the big data age,data has grown explosively,constantly updated,and has a complex structure.The traditional comprehensive evaluation has shown two shortcomings.First,it is difficult to model when there are too many parameters and too complex structure.Second,the accuracy of the model is getting lower and lower with the update of the data.In order to improve this situation,this paper attempts to adopt the self-adaptive ensemble learning Adaboost perspective in statistical learning to develop a data-driven evaluation model that is “adaptive” to data as it updates.The basic idea of data-driven evaluation is to improve and integrate the evaluation model based on continuously updated data,so that it can continuously adapt to data flow and improve model accuracy.Based on this concept,this dissertation mainly does two aspects of work.First,the Topsis method commonly used in the multi-attribute decision making evaluation method is selected,and a parameter ? is added to the scoring formula to improve the Topsis method,which can be changed with the data.Second,It is noted that an evaluation model cannot adapt to today's complex data structures.The principal components and factor analysis methods that adapt to multiple indicators are considered together with the data-driven Topsis method as the weak evaluation models.Then the integration is integrated from the AdaBoost perspective to achieve the final integration:Evaluation model.Finally,in order to test the effect of the evaluation models constructed,this paper applies them to the influence evaluation of WeChat Subscription.The results show that,compared with the traditional Topsis method,the data-driven Topsis method reduces the error rate by 37.14%;the integrated evaluation model has a correct rate of 88.57%,which is 54.28%,17.14%,31.43%,28.57% and 62.86% higher respectively compared with the Topsis method,data-driven Topsis method,principal component method,factor analysis method and average value method. |
PDF Full Text Request