社团演化特征构造及预测

Feature Construction and Prediction of Community Evolution

现实世界中的网络往往会随时间推移逐渐改变,社团演化预测通过分析动态网络数据判断社团的发展趋势,对于理解复杂网络演化规律及其应用具有重要意义.社团演化特征构造从历史数据中提取社团结构、时序特征用于预测,其是否准确刻画社团特性直接影响预测结果准确率,是研究中的关键问题.本文提出一种基于多元特征构造的社团演化预测方法,从动态网络中提取社团的结构(微观、介观、宏观)、时序、行为特征,并采取针对多重长度演化链的集成方法进行分类.在两类实际数据集上进行的实验表明了该方法预测准确性优于已有研究.

Myriad real-world network systems have actors and interactions change as a function of time. Predicting the future trend of community based on the analysis of dynamic network data is a problem with high theoretical and practical significance. Constructing structural and temporal features of communities from the history data plays a key role in the community evolution prediction. The precision of result relies significantly on how well the measures describe the community profile. In this paper,we proposed a community evolution prediction approach based on multiple features construction. Our approach extracts both structural（ microscopic,mesoscopic,and macroscopic） features,temporal features,and behaviour features of communities,and in consequence as input of a modified ensemble classifier for multi-length evolution chains. Comparison to previous studies on two types of real-world networks shows that improve predictions.