报告题目：Unexpectedly high capacity to spread information and extremely imbalanced discursive power of social media networks

摘要：Online social networks have emerged as an important medium for the spread of information and have been used in various fields, from political campaign to marketing, disaster relief and social sensing. All of these applications rely on how information spreads on social media networks. Most studies assume that information spreading is a percolation process and large cascades occur only when the retweet probability of information items exceeds the percolation critical point, also known as tipping point. However, whether this widely used hypothesis is valid in current large-scale social media remains unclear. Here we continuously observe at least 0.18 million users’ online behaviors for three years in Weibo, the biggest microblog social medium in China, crawl almost the whole friendship network of 100 million users and collect a large number of information tracks within the same period of time. We find that the cascading threshold is only one tenth of that theoretically obtained previously, and 98.4% of the information items that have led to outbreaks in real social media could be incorrectly predicted to be at subcritical states by the existing theories. This finding indicates that the capacity of social media to spread information has been seriously underestimated. Moreover, the positive-feedback effect in the coevolution between user activity and net-work structure, on both Weibo and Twitter, becomes stronger with time. Such a stronger effect induces extreme imbalance in users’ discursive power. Indeed, we find that the top 0.7% of users possess 99.3% of discursive power, 17 times more serious than previous theoretical prediction. We incorporate the coevolution mechanism into network percolation theory, offering a novel model that agrees with empirical data much better than previous ones. Taken together, our results deepen the understanding of phase transition and coevolution dynamics in social media, applicable to a wide range of problems pertaining to information cascades on networks.

报告人：胡延庆

博士，现在为中山大学数据与计算机学院副教授（中大百人计划），博士生导师。2011年毕业于北京师范大学系统科学学院，获得系统理论方向理学博士学位，并获得北京市优秀博士论文奖；2011-2013年纽约城市大学Levich Institute 博士后。近几年主要从事具有图或者网络结构的大数据挖掘与人工智能算法与理论研究工作，探索数据背后的自然物理规律。发表论文 40 余篇，其中通讯、第一作者论文 25 篇，包括 Nature Physics, PRL, PRX 各 1 篇，PNAS 2 篇，PRE 11 篇，其中 Nature Physics 论文被选入该期封面推荐论文。

时间：2019年10月28日 下午15:00

地点：中心校区知新楼B座1032报告厅

邀请人：王光辉 数学学院教授