A Gradient Tree Boosting based Approach to Rumor Detecting on Sina Weibo

Abstract

Rumor detecting on microblogging platforms such as Sina Weibo is a crucial issue. Most existing rumor detecting algorithms require a lot of propagation data for model training, thus they do not have good detecting accuracy at the early stage after a rumor message is posted. In this paper, we propose to use gradient tree boosting (GTB) approach to rumor detecting, based on which a rumor detecting algorithm is developed. At the same time, the GTB-based approach makes it easy to conduct feature selection, and a feature selection algorithm is developed. Experiments on a widely used dataset of Sina Weibo show that the proposed detecting algorithm outperforms the state-of-the-art detecting algorithms; moreover, it has the highest detecting accuracy at the early stage. This work seems to be the first to use GTB-based approach in rumor detecting, and the results suggest that it may be a promising one.