| Social spam detection is a binary classification problem, Using binary classification of SVM support vector machine can be achieved it. There are many problems on social spam detection using traditional SVM:The scale of social spam is very large, and its data has high dimensional and complex characteristics. It takes a long time and affects the learning precision using traditional SVM to train and learn social spam data directly. This article improve the traditional SVM using semi-definite programming kernel. The main work is as follows:1. Use K-means clustering algorithm to extract characteristic vectors, it is a method to improve the efficiency of SVM training. Social spam data sets have strong nonlinear, the euclidean distance of the traditional K-means clustering algorithm is unable to deal with nonlinear data effectively, the quality of the social spam characteristic vectors is not high. SVM kernel function has a stronger nonlinear mapping capability, this article integrate kernel function and K-means clustering algorithm, namely semi-definite programming kernel K-means clustering algorithm. And improve the capacity of extracting social spam using K-means clustering algorithm. Then through the experiment to prove the feasibility and effectiveness of the new algorithm.2.Different SVM kernel functions have different nonlinear mapping features. one way to improve the ability of SVM classification is using the method of composing various of kernel function. Therefore, this article proposes to use semi-definite programming method to determine the best combination coefficient of the kernel function, to structure semi-definite programming kernel SVM. So the author propose the new algorithm "research on folksonomy social spam detection algorithm based on semi-definite programming kernel". The experiments show that this algorithm has a good effect on social spam classification, it is not only improved the classification accuracy, but also reduced the training time. |
PDF Full Text Request