“…Specifically, some methods provide in-depth analysis around content features, e.g., linguistic (Reis et al, 2019), semantic (De Sarkar et al, 2018), emotional (Ajao et al, 2019), and stylistic (Potthast et al, 2018), and achieve limited performance. On this basis, some work additionally extracts various social context features as credibility features, including meta-data based (i.e., source-based (Rathore et al, 2017;Yu et al, 2018), user-centered (Long et al, 2017;Ribeiro et al, 2017), and post-based (Wang, 2017;Ma et al, 2018b)) and networkbased (Ruchansky et al, 2017;Liu & Wu, 2018;, and promotes the development of different fusion approaches, such as hybrid-CNN model (Wang, 2017), CSI model (Ruchansky et al, 2017), and tree-structured RNN (Ma et al, 2018b), which gain remarkable performance boosts compared to other models only capturing text features. From these methods, we can find that expanding features can significantly improve the performance of credibility evaluation.…”