EasyLink

EasyLink is an Easy-to-use collection of SOTA Link Prediction models for Networks, especially Social Networks. It is designed to provide tutorials for beginers, pipeline implementation of models and comprehensive guidance for choosing the right model.

Requirements

• PyTorch>=1.13.0
• PyTorch_Geometric
• OGB>=1.3.1

Models

Model	Paper	Module
Common Neighbors	[2003] The Link Prediction Problem for Social Networks	Model /Example
Adamic Adar	[2003] Friends and neighbors on the Web	Model /Example
Resource Allocation	[2009] Predicting missing links via local information	Model /Example
Local Path Index	[2009] Similarity index based on local paths for link prediction of complex networks	Model /Example
Node2Vec	[KDD2016] node2vec: Scalable Feature Learning for Networks	Model
GraphSage	[NIPS2017] Inductive Representation Learning on Large Graphs	Model
SEAL	[NIPS2018] Link prediction based on graph neural networks	Model /Example
NBFNet	[NIPS2021] Neural Bellman-Ford Networks: A General Graph Neural Network Framework for Link Prediction	Model /Example

DataSets

DataSet	Description	Statistics	Node Fea	Edge Fea
USAir	a network of US Air lines	332 nodes and 2,126 edges	-	-
Facebook	sampled friendship network of Facebook.	4,039 nodes and 88,234 edges	-	-
Arxiv	collaboration network generated from arXiv	18,722 nodes and 198,110 edges	-	-
twitter	social circles from Twitter	81,306 nodes and 1,342,310 edges	-	-
MAG-collab	subset of the collaboration network between authors from MAG	235,868 nodes and 1,285,465 edges	Yes	-
CORA	subset of the paper citation network of machine learning	1433 nodes and 2708 edges	Yes	-

RoadMap

• Build classic heuristic link prediction methods: Common Neighbors, Adamic-Adar, Resource Allocation
• Build basic graph neural network models: Node2Vec, GraphSage
• Build advanced graph neural network models: SEAL, NBFNet
• Demonstrate examples for each model
• Import classic link prediction datasets: USAir, Facebook, Arxiv, twitter
• Develop a universal pipeline for link prediction
• Optimize computational efficiency
• Host 1~3 leaderboards on diverse datasets with link prediction models
• Launch a tutorial website with complimentary documentation
• Support a easy-to-use python package
• Explore an automatic pipeline for link prediction