A new approach for detecting spam microblogs based on text and user's social network features

Abstract

Recently more and more spam messages are emerging on microblogs, which leads to an unpleasant or even deteriorating social network environment. Existing studies on spam microblog detection mostly make use of textual features or social network features alone to detect spam messages. While in this paper, we propose a new detection approach from users' perspective, which combines social network features of the publishers with textual features of microblogs itself together, to compose a feature vector. By feeding the feature vector into a SVM machine learning system for data training, we classify spam microblogs from benign ones. We conduct experiments with the dataset of Sina Weibo, one of the most famous Chinese microblogs, to verify the effectiveness of our approach. Compared to the approaches which only consider textual or network features, we observe 13% and 29% increases of accuracy respectively with our proposed approach.