Line graph contrastive learning for link prediction

Link prediction tasks focus on predicting possible future connections. Most existing researches measure the likelihood of links by different similarity scores node pairs and predict between nodes. However, similarity-based approaches have some challenges in information loss nodes generalization ability indexes. To address above issues, we propose a Line Graph Contrastive Learning (LGCL) method to obtain rich with multiple perspectives. LGCL obtains subgraph view h-hop sampling target pairs. After transforming sampled into line graph, link task is converted classification task, which graph convolution progress can learn edge embeddings from graphs more effectively. Then design novel cross-scale contrastive learning framework maximize mutual them, so that fuses structure feature information. The experimental results demonstrate proposed outperforms state-of-the-art methods has better performance robustness.