Research And Application Of Feature Combination Recommendation Model Based On Modular Neural Network

With the advent of the era of big data,it is an urgent need to improve the cognitive ability of large-scale data.In view of the problems of single artificial neural network in the field of cognitive computing,such as large structure,poor explanation and high time complexity,this paper introduces the modularity of brain "function separation" into single neural network to build modular neural network to improve the ability of artificial neural network to process complex information.This paper proposes a feature combination recommendation model based on modular neural network,which is used in the credit scoring system of bank.(1)In this paper,the optimal K-mean density clustering K-DB algorithm is proposed as the module partition method of modular neural network.In order to improve the accuracy of module partition,K-DB algorithm combines two strategies to get better module partition effect.(2)According to the needs of feature extraction for massive data in the era of big data,the characteristics of competitive learning mechanism of SOM neural network are studied.In view of the situation that neural network weights are easily adjusted into local optimal solution,sa-som neural network is optimized by simulated annealing algorithm as a model subnet to improve the performance of a single SOM neural network.Through experiments,competitive learning is verified The efficiency of feature sorting.(3)The density clustering K-DB algorithm for optimal K-means is designed to divide the modules.Sa-som network is selected as the sub network for feature extraction,and the membership based fuzzy reasoning method is added to the modular neural network MNN model for feature combination output.In the model,K-DB algorithm is used to cluster the input sample space horizontally to get the core points in the cluster,and sa-som sub neural network is used to extract the vertical features of the core points to get the importance order of the core point attributes.(4)This paper designs a bank credit evaluation system based on MNN model toachieve fast credit rating of bank credit customers,improve the accuracy of bank credit ability evaluation of customers and improve the work efficiency of bank credit business personnel.According to the results of MNN model extraction,combined with the bank's open user behavior information scoring standards,the system makes a comprehensive quantitative scoring for bank credit users.By calculating the PSI index,KS value and AR value of the model,it is verified that the model has high stability and the ability to distinguish good and bad accounts.Compared with the accuracy of traditional credit scoring methods,the experimental results show that the credit scoring method based on MNN model has better performance.