Link Prediction in Signed Social Networks

Signed social networks extend traditional network models by incorporating both positive and negative links to represent nuanced relationships such as trust versus distrust or friendship versus hostility. Link prediction in this context seeks to infer not only whether a connection will form but also its polarity, a task that underpins recommendation systems, fraud detection, community discovery and the analysis of ideological or sentiment flows. Early methods drew on sociological principles such as structural balance theory and status theory to generate heuristic scores for missing edges. More recent approaches employ latent‐factor models, matrix factorisation and graph neural networks to learn node embeddings that capture complex structural patterns. Key challenges include the sparsity and imbalance of negative links, dynamic changes in relationships and the integration of local triadic configurations with global topology. State-of-the-art techniques now combine contrastive or multi-view learning, random-walk diffusion tailored to signed graphs and hierarchical pooling to produce end-to-end representations that improve both sign and existence prediction, offering greater robustness and interpretability for signed network analysis.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Link Prediction in Signed Social Networks publication trend

The graph below shows the total number of articles in link prediction in signed social networks across all publications each year (not limited to Nature Index journals).

Technical terms

Signed social network: A graph in which edges carry positive or negative labels to indicate the nature of relationships between nodes.

Link prediction: The task of estimating the likelihood and type of a future connection between two nodes in a network.

Structural balance theory: A sociological model describing how configurations of signed triangles influence network stability and tension.

Random-walk diffusion: A process whereby node features propagate through a network via probabilistic walks, specially adapted to account for edge signs.

Graph neural network (GNN): A deep learning architecture that learns node representations by aggregating information from neighbours according to the graph topology.

References

1. A Community‐Based Approach for Link Prediction in Signed Social Networks. Scientific Programming (2015).
2. Signed random walk diffusion for effective representation learning in signed graphs. PLOS ONE (2022).
3. Learning Embedding for Signed Network in Social Media with Hierarchical Graph Pooling. Applied Sciences (2022).
4. Learning Weight Signed Network Embedding with Graph Neural Networks. Data Science and Engineering (2023).