Encoding Following Information with Graph Attention Network for Opinion Prediction

Opinion prediction is an important and challenging task. Previous work often only predicts opinions based on users’ content, ignoring the influence of neighbors’ opinions on users in social networks. In this paper, we utlize Graph Attention Networks to capture neighbor information for opinion prediction. Firstly, we divide users in the online social media platform into general users and online influencers by the number of their fans. Then we built the implicit link between general users by their following list, and construct the implicit network for opinion prediction. Experimental results show that the proposed model is more accurate in opinion prediction compared to many competitive baselines.