Link Prediction¶

Algorithms for predicting missing or future edges.

NetworkX Integration

Link prediction algorithms are available through the NetworkX adapter.

Using the NetworkX Adapter¶

import grafeo
import networkx as nx

db = grafeo.GrafeoDB()

# Convert to NetworkX graph
nx_adapter = db.as_networkx(directed=False)
nx_graph = nx_adapter.to_networkx()

# Get non-edges for prediction
non_edges = list(nx.non_edges(nx_graph))[:100]  # Sample

Common Neighbors¶

Predict links based on shared connections.

preds = nx.common_neighbor_centrality(nx_graph, non_edges)
for u, v, score in preds:
    print(f"({u}, {v}): {score:.4f}")

Jaccard Coefficient¶

Normalized common neighbors.

preds = nx.jaccard_coefficient(nx_graph, non_edges)
for u, v, score in preds:
    print(f"Jaccard({u}, {v}) = {score:.4f}")

Adamic-Adar Index¶

Weighted common neighbors (rare neighbors count more).

preds = nx.adamic_adar_index(nx_graph, non_edges)
for u, v, score in preds:
    print(f"Adamic-Adar({u}, {v}) = {score:.4f}")

Preferential Attachment¶

Product of node degrees.

preds = nx.preferential_attachment(nx_graph, non_edges)
for u, v, score in preds:
    print(f"PA({u}, {v}) = {score}")

Resource Allocation¶

Similar to Adamic-Adar but with different weighting.

preds = nx.resource_allocation_index(nx_graph, non_edges)
for u, v, score in preds:
    print(f"RA({u}, {v}) = {score:.4f}")

Use Cases¶

Friend suggestions in social networks
Product recommendations
Knowledge graph completion
Collaboration prediction