Link prediction is the machine-learning task of inferring missing or future edges in a graph from its observed structure and node attributes. Techniques range from similarity heuristics and matrix factorisation to graph neural networks and knowledge-graph embeddings. It underpins recommendation, knowledge-graph completion, and social-network analysis.
Content
- Embedding methods such as TransE and graph neural networks learn latent node representations whose geometry encodes the likelihood of a connection. Evaluation uses ranking metrics like mean reciprocal rank and Hits@k against held-out edges, balancing precision against the sparsity of true links.