Detection Splog Algorithm Based on Features Relation Tree

Blogosphere has become a hot research field in recent years. As the existing detection algorithm has problems of inefficient feature selection and weak correlation, we propose an algorithm of splog detection based on features relation tree. We could construct the tree according to the correlation of the features, reserving the strong relevance features and removing the weak ones, then prune the redundant and irrelevance features by using the secondary features selection method and retain the best feature subset. The experimental results conducted in the Libsvm platform show that the algorithm based on the features of relation tree has higher precision and covering rate compared to the traditional ones. The precision of the algorithm on simulated training remains at about 90%, which has better generalization ability.