Graph-Based Interactive Matching for Pairs of News Articles

Kunhao Pan, Guowei Zhang, Meng Liao, Jin Xu

Abstract: Long-text document matching has been widely applied in many applications, such as topic detection and tracking and relative article recommendation. However, existing methods still have many defects in extracting and utilizing long text features, especially in news articles. In this paper, we propose a novel long-text pair matching framework that constructs texts into graphs and comprehensively utilizes graphs for interactive matching. We conduct extensive experiments on four datasets, including CNSE, CNSS, TNSE, and TNSS. Extensive experimental results demonstrate the significant improvements over a wide range of state-of-the-art methods. The proposed EEG model is novel, and it significantly outperforms an extensive range of baselines.