中文摘要：

有效预测社交网络中的社区演化规律和趋势,在广告精准投放、网络舆论监测与引导方面具有广泛的应用前景.近年来,基于事件框架的社区进化预测建模在反映社区演化规律和趋势方面一直是热点,研究的难点在于提高预测模型的预测准确度.为解决这一难点,文中首先提出一种改进的事件框架,以新型事件框架为基础,针对不同事件分别构建有效的预测模型.其中预测模型的输入指标包括网络结构特征、社区的结构特征和社区进化特征.最后为了充分验证所建预测模型的有效性并保证预测模型的实用价值,在实验中分别用人造动态网络数据集、DBLP动态网络数据集和Facebook数据集作为实验数据集,以确保实验数据集的多样性.实验结果表明文中的预测模型针对形成、消失、保持、合并和分裂等事件的预测具有较高的准确性,结果也表明文中提出的特征指标和对应事件的预测模型在实际事件预测中将具有较高的实用价值.

英文摘要：

Research on community detection is one of the fundamental investigations in social networks,and event-based frameworks for characterizing the community evolution and revealing the tendency of communities in social networks can be lately seen in some international workshops,conferences and seminars.To improve the predictive accuracy of predictive models based on event frameworks is a difficult part for present research,therefore,in this paper we proposed an improved event-based framework.According to the improved definition of all kinds of events,we presented some effective predictive models which have many important applications,such as recommending advertisements and guiding public opinions.To find appropriate predictive models which are able to predict community events with higher accuracy,we presented several features involved social network structure,community structure and community evolution features.In our experiments,we use real world datasets including artificial dataset,DBLP datasets and Facebook datasets which are classical and often used datasets to validate the efficiency of our predictive models.The experimental results confirmed that the predictive models with features presented by us can predict the evolution of communities including birth,death,remain,merge and split accurately,and the results also show that the predictive models would have high practical value.