Abstract:
Prediction of links’ sentiment tendency based on the analysis of the emotional relationship between nodes in information networks is widely used in commercial marketing, content recommendation, and other fields, which is a research focus of network analysis. The traditional studies on the prediction of links’ sentiment didn’t mine the information in data sufficiently. They often ignore the deep semantics of data or the node attributes and so on. Therefore, the accuracy of prediction can be further improved. For the above problems, we propose a model for predicting links’ sentiment tendency in this paper, which integrates multiple types of information in heterogeneous networks. The model introduces a basic estimate as a rough evaluation of the emotional relationship between specific nodes and combines information from similar nodes as well as nodes’ attributes to complete the prediction task. When capturing nodes with similar sentiment tendencies in the network for prediction tasks, a traversal walk method based on meta-path with restricted path types was proposed. Extensive experiments on five real-world datasets demonstrate the effectiveness of the proposed model. Moreover, the experimental results show the capability of the proposed model to deal with sparse matrix and cold-start problems and reveal the efficacy of each component of the model throughout the prediction process.