基于复杂网络的微吧话题流行度预测研究

Popularity Prediction of Micro-bar Topic Based on Complex Network

微吧是微博平台的贴吧,具有良好的话题主题性,为更好地进行话题流行度预测研究,对电影吧作为研究对象,建立复杂网络模型,并设计自适应差分进化算法对网络进行训练。在算法参数的设计中,通过对算法复杂度边界的推导得到最优参数。对比实验表明,较于神经网络该模型结构更加灵活,较于拟合模型该模型能够对话题特性进行更为全面的描述,并且在话题流行度的预测中具有更高的准确度和稳定性。

Micro-bar is similar to post bar on mircoblog platform, which is based on subjects. To study the features of topic, movie bar was chose to be the research object. A model of complex network was established with topic popularity factors abstracted by correlation analysis. Self-adaptive differential evolution algorithm has been applied to train the network. In the design of the algorithm, the main parameters were set through complexity theoretical derivation to reduce complexity. In contrast, this model is more flexible than neural network and is able to describe more topic characteristics than fitting model, but also apply to forecasting the popular topics with higher accuracy and stability.