Name
Abstract | ||
|---|---|---|
Predicting the popularity of a topic is an important task for modeling and analyzing the trending patterns of social network topics. Deep learning-based methods are one of the mainstream approaches to deal with this task. Among them, the propagation structure is used to assist in predicting topic popularity, which can well model the evolution process of propagation. However, existing methods essentially predict the sampled value of a topic on the sampled data, which is different from the recorded value in the real world. In this paper, we mainly complete the task that uses the sampled data and historical real values to predict the future real forwarding value. We propose a temporal analysis model based on Long Short-Term Memory (LSTM) for historical real values and apply a Graph Convolutional Network (GCN) on the sampled data to extract the spatial features of propagation to assist in final prediction. To reduce the loss caused by using sampled features to predict the true value, we design a sub-module termed debias regressor that corrects bias after feature extraction. Passing the original predicted value through the regression debiasing sub-module can make the predicted value better fit the true records. Experimental results show that our proposed method has a great improvement in reality prediction compared with other popularity prediction methods, and is of great help in practical applications. |
Year | DOI | Venue |
|---|---|---|
2022 | 10.1109/DSC55868.2022.00072 | 2022 7th IEEE International Conference on Data Science in Cyberspace (DSC) |
Keywords | DocType | ISBN |
Sample Debias,Popularity Prediction,Information Diffusion | Conference | 978-1-6654-7481-8 |
Citations | PageRank | References |
0 | 0.34 | 7 |
Authors | ||
4 | ||