Retweet Prediction Using Context-Aware Coupled Matrix-Tensor Factorization

摘要

Retweet behavior plays an important role in the process of information diffusion on social networks. Although many researches have been studied the problem of retweet prediction, these studies ignore the important characteristic of multiple contextual dimensions for user's decision in the modeling process. To this end, we propose a novel multiple dimensions retweet prediction model based on context-aware coupled matrix-tensor factorization (RCMTF). This model first introduces a reference tensor based on the historical retweet behavior patterns to alleviate the problem of data sparsity, and then constructs three contextual factor matrices from user and message and influence dimensions on basis of network structure, message content and historical interactions to further improve the prediction accuracy. Finally, we collaboratively factorizes these contextual factors under matrix and tensor factorization models framework for predicting user's retweet behaviors. Extensive experiments are conducted to demonstrate the effectiveness of the proposed method on two real-world datasets. The results show that our proposed model outperforms the state-of-the-art methods.