Modeling Temporal Interaction for Dynamic Sentiment Analysis on Social Network

With the evolution of the network over time, dynamic sentiment is ubiquitous in the real social network. Learning the temporal social interactions representation and modeling a dynamic socio-sentiment analysis model is important to understand the network data and necessary for accurately analyzing and prediction. In this work, we design a temporal social network representation model for dynamic sentiment analysis by capturing the temporal interaction information in the evolutionary social network. Specifically, the temporal social network embedding method is employed to learn dynamic representations of node and node's interaction relationships from the evolutionary network by preserving both the explicit structural proximity information and implicit multiview association information. The joint temporal heterogeneous social network embeddings are learned by fusing the different dimensional representation at their temporal granularity, which can be used to naturally support sentiment analysis on social network in a dynamic way. The results demonstrate that the raised approach reports better performance than the baseline methods. The results indicate the importance of incorporating temporal dependencies in social network for dynamic sentiment analysis. It also indicates the effectiveness of the proposed approach for learning meaningful dynamic network representations to improve sentiment analysis performance.