Research And Implementation Of Online Service Reputation Measurement Based On Semi-supervised Learning

Online service refers to an online transaction mode that provides services to users through Internet technology.It has been rapidly developed and well received by users because of its convenient use,simple operation and low cost.However,with the rapid development of online services,there are also many problems.For example,in the face of a large number of services with the same functions and different quality of service,it is difficult for users to make correct service choices.In addition,in view of the particularity of online trading mode,users can not try online services,so that users are faced with huge transaction fraud risk.Credit is a comprehensive measure of service quality,which can truly reflect the extent to which service providers fulfill their commitments.In the current credit measurement research,the model based on user feedback score does not consider the impact of service dimension attributes on service reputation,while the reputation measurement model based on multi-dimensional evaluation information often ignores the impact of service attributes on service reputation.Therefore,it is particularly important to study a service reputation measurement method,which can objectively and effectively help users make correct service choices and deal with transaction fraud.To solve the above problems,this paper adopts an online service reputation measurement method which integrates multi-dimensional attribute information of services.Essentially,service reputation measurement can be regarded as the classification of different types of service reputation.Therefore,service reputation measurement problem is modeled as the classification of service reputation.Service reputation classification is realized by training service classifier model,and then service reputation measurement is realized.In this paper,a distance-constrained semi-supervised learning method is used to train the service classifier model based on Tri-training semi-supervised learning.Firstly,a small number of service samples are labeled manually as the initial training set.Through playback sampling of service samples in the initial training set,several training sets with different distribution of service samples are obtained and several different decision tree classifications are trained.Device;Then,different decision tree classifiers are used to classify and predict the unlabeled centralized services,and the predicted services and prediction labels are added to the training set to retrain the classifier,so that the classifier model can be trained adequately by using a small number of labeled services and a large number of unlabeled services;finally,the unknown online classifier is trained based on the final training classifier.Service reputation classification is used to measure online servicereputation.This paper validates the rationality of service reputation measurement by modeling service classifier model based on related theory and technology,and validates the validity and efficiency of the model in classification accuracy,recall rate and F value by using real online service data sets.Finally,a prototype system of online service reputation measurement is designed according to the method theory applied in this paper,which can realize the service reputation measurement.