KAPP: Knowledge-Aware Hierarchical Attention Network for Popularity Prediction

Abstract

AbstractWith the social networks becoming a major source of information in recent years, predicting the popularity of information in social networks has appeared intriguing to researchers in both academia and industry. However, existing methods still lack the utilization of external knowledge features, and are hard to extract the internal knowledge correlation of social information.In this paper, we propose a knowledge-aware hierarchical attention network for popularity prediction (KAPP), which integrates the representation of the knowledge graph into popularity prediction. We aim to learn information representation based on temporal point process and knowledge graphs simultaneously from social content. In information cascading, we design a hierarchical attention mechanism to simulate the attention of human beings and the influences of users in social networks, and naturally establish the model structure from knowledge characteristics to popularity prediction.In All, through attention mechanism for knowledge graph expression and analogy learning of temporal point process, our work makes an efficient prediction for information in social networks with the deep learning method. On the real-world data set of Weibo, we evaluate our model with intensive experiments and metrics, which outperforms previous methods including traditional approaches and deep learning methods.