Social Network Development Prediction System Using Supervised Classifiers Hybrid Method

Social networks are a new generation of websites that are in the focus of the users of the global Internet these days. In such networks, predicting the occurrence of links is a basic and fundamental problem that is related to the probability of the existence of a link between two network nodes that are not connected. In this research, in order to predict this issue, the method of combining classifiers and the combined ECOC algorithm has been used. Also, in order to implement the mentioned method, the main clustering algorithm, which is a support vector machine, was executed several times on the data set, and each time only the order of entering the data into the algorithm was changed. With this action, the effect of how the data is placed in the output was considered. At the end, the output for the final evaluation or the output with the most repetition was selected by the majority voting method. Also, in order to compare the test results with other methods, the results obtained from the average of ۵۰ independent executions of the program were compared with the results of several well-known and widely used algorithms in this field, such as: KNN, Random Forest, Bagging, Decision Tree, Multilayer-Perceptron algorithms. The result of this comparison proved that the proposed method has provided better results than other methods and different categories.

کلیدواژه ها:

machine learning ، Supervised Classifiers ، big data problems ، link prediction