Social media has become an important channel for publicizing academic research. Employing a dataset of about 10 million tweets of 584,264 scientific papers from 2012 to 2018, this study investigates the differential diffusion of influential and non-influential papers (divided by Average journal impact factor percentile). We find that non-influential papers shows a diffusion trend with multiple rounds, sparse, short-duration and small-scale bursts. In contrast, the bursts of influential journals are characterized by a small number of persistent, dense and large-scale bursts. Influential papers are generally disseminated to many loosely connected communities, while non-influential papers are diffused to several densely connected communities.