Tracking Dynamic Magnet Communities: Insights from a Network Perspective

Communities, such as user groups, companies and countries, are important objects in social systems. Recently, researchers have proposed numerous quantitative indicators to measure the attractiveness of a community. However, most of these indicators are mainly under static settings and lack the predictive power of future impact/attractiveness. Meanwhile, in many real-world applications, especially in finance, it is of great interest for the stakeholders to identify the communities, not necessarily the most influential ones at the moment, but the future leaders for years to come. Given the increasing availability of entity-community interaction evolution records, it’s natural to exploit them to model the network changes of communities. We refer the change of community interaction as attention flow and define communities that will sustainably attract more entities’ attentions than others in a future time interval as dynamic magnet communities. We study the problem of dynamic magnet community identification based on entity-community interaction evolution records. Two major challenges are identified as follows: (1) temporal dynamics, it’s difficult to model the rising-declining trend of interactions; (2) sustainability constraints, the effect of attention flow on community prosperity is complex, where too rapid attention growth increases the corruption risks. In response, we propose to model the interaction network evolution of different communities over time by lasso based growth curve fitting. Taking sustainable attention flow into account, we measure attention flow utility from benefit and risk perspectives, and further present a hybrid approach of local and global ranking to track dynamic magnet communities. Due to the lack of dataset for testing, we collected a dataset of international business merger and acquisition network among different countries in the world. The experimental results demonstrate the effectiveness of our proposed model.