Prediction of popularity has profound impact for social media, since itoffers opportunities to reveal individual preference and public attention fromevolutionary social systems. Previous research, although achieves promisingresults, neglects one distinctive characteristic of social data, i.e.,sequentiality. For example, the popularity of online content is generated overtime with sequential post streams of social media. To investigate thesequential prediction of popularity, we propose a novel prediction frameworkcalled Deep Temporal Context Networks (DTCN) by incorporating both temporalcontext and temporal attention into account. Our DTCN contains three maincomponents, from embedding, learning to predicting. With a joint embeddingnetwork, we obtain a unified deep representation of multi-modal user-post datain a common embedding space. Then, based on the embedded data sequence overtime, temporal context learning attempts to recurrently learn two adaptivetemporal contexts for sequential popularity. Finally, a novel temporalattention is designed to predict new popularity (the popularity of a newuser-post pair) with temporal coherence across multiple time-scales.Experiments on our released image dataset with about 600K Flickr photosdemonstrate that DTCN outperforms state-of-the-art deep prediction algorithms,with an average of 21.51% relative performance improvement in the popularityprediction (Spearman Ranking Correlation).
展开▼
