Community evolution prediction based on feature change patterns in social networks
Abstract Predicting community evolution in dynamic social networks is crucial for relevant authorities to understand trends and implement safety measures in advance. Most existing algorithms for predicting community evolution rely on extracting community state features to forecast evolutionary event...
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2025-04-01
|
| Series: | Scientific Reports |
| Subjects: | |
| Online Access: | https://doi.org/10.1038/s41598-025-91766-7 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Predicting community evolution in dynamic social networks is crucial for relevant authorities to understand trends and implement safety measures in advance. Most existing algorithms for predicting community evolution rely on extracting community state features to forecast evolutionary events. However, in highly interactive social networks, such as corporate collaboration networks in financial markets, extracting high-quality community state features is extremely challenging. This study proposes a community evolution prediction method based on feature change patterns, aiming to explore the changing features during community evolution, and designs an algorithm to learn the rules of feature changes, thereby obtaining the feature change pattern of the community. Compared to traditional methods that rely on static state features, our proposed approach captures richer dynamic information and more accurately reflects community evolution trends. Additionally, we have designed a parallel learning strategy with parameter sharing, based on the consistency of community environments. Experimental results show that our method, based on feature change patterns, achieves approximately 25% improvement in maximum predictive performance on the AS, DBLP, and Facebook datasets compared to baseline methods (TNSEP, GNAN, and MF-PSF). Additionally, the parallel learning mechanism reduces training time by nearly half. |
|---|---|
| ISSN: | 2045-2322 |