Data for: AFIF: Automatically Finding Important Features in Community Evolution Prediction for Dynamic Social Networks

Networks
DBLP network is divided into eleven time windows (time span 01/01/2003 to 31/12/2013).
Facebook Wall Posts network is divided into eight time windows (time span 01/01/2005 to 31/12/2008).
Wiki-Talk network is segmented into six time windows (time span 24/11/2007 to 31/12/2007).
Enron email network is segmented into twelve time windows (time span 01/01/2001 to 31/12/2001).
Reddit-reply network is segmented into six time windows (time span 07/01/2014 to 13/01/2014).
Stack Overflow network is segmented into six time windows (time span 24/01/2016 to 29/02/2016).

Social group discovery
Communities of each time window are discovered using Infomap, Label Propagation, and Leiden algorithms.
For running the community detection algorithms, we assume that the networks are undirected and unweighted graphs.
The communities whose size was smaller than two members were ignored.

Community evolution tracking and chain identification
In order to track community evolution, we investigate each community to find its similar community or communities from previous time windows, which is called community matching.
We employed ICEM (Identification of Community Evolution by Mapping) method in order to determine the evolution events because it is a highly efficient approach to track community evolution and considers partial evolution and non-consecutive time windows (Kadkhoda Mohammadmosaferi & Naderi, 2020). ICEM has two parameters which are α and β, in this paper, the thresholds for being partially similar and very similar are set to α=10% and β=90%, respectively.

Each uploaded Dataset contains chains of evolution for a network and a community detection algorithm.

Reference:
Kadkhoda Mohammadmosaferi, K., Naderi, H., 2020. Evolution of communities in dynamic social networks: An efficient map-based approach. Expert Syst. Appl. 147, 113221. https://doi.org/10.1016/j.eswa.2020.113221

网络

DBLP网络被划分为11个时间窗口，时间跨度为2003年1月1日至2013年12月31日。
Facebook墙帖网络被划分为8个时间窗口，时间跨度为2005年1月1日至2008年12月31日。
Wiki-Talk网络被划分为6个时间窗口，时间跨度为2007年11月24日至2007年12月31日。
Enron电子邮件网络被划分为12个时间窗口，时间跨度为2001年1月1日至2001年12月31日。
Reddit回复网络被划分为6个时间窗口，时间跨度为2014年1月7日至2014年1月13日。
Stack Overflow网络被划分为6个时间窗口，时间跨度为2016年1月24日至2016年2月29日。

社交群体发现

针对每个时间窗口的网络，采用Infomap、标签传播（Label Propagation）以及Leiden算法开展社区发现研究。在运行社区检测算法时，我们假设网络为无向无权图。同时，将成员数量小于2的社区予以剔除。

社区演化追踪与链识别

为追踪社区演化过程，我们对每个社区进行匹配操作，即从前序时间窗口中查找与其相似的一个或多个社区，该过程称为社区匹配。本文采用ICEM（基于映射的社区演化识别，Identification of Community Evolution by Mapping）方法来确定演化事件，因其是一种高效的社区演化追踪方案，可兼顾部分演化与非连续时间窗口场景（Kadkhoda Mohammadmosaferi & Naderi, 2020）。ICEM包含α和β两个参数，本文将部分相似与高度相似的阈值分别设置为α=10%与β=90%。

每个上传的数据集均包含对应网络与社区检测算法对应的演化链。

参考文献：
Kadkhoda Mohammadmosaferi, K., Naderi, H., 2020. 动态社交网络中的社区演化：一种高效的基于映射的方法. Expert Syst. Appl. 147, 113221. https://doi.org/10.1016/j.eswa.2020.113221