Metagraph2vec

Complex Semantic Path Augmented Heterogeneous Network Embedding

Code Review

BEST FOR You�ORGANICS COMPANY

Methodologies

A new method for heterogeneous network embedding that learns more informative embeddings by capturing richer semantic relations between distant nodes.

• Present metagraph-guided random walk to generate heterogeneous neighborhood in an HIN.
• Present the MetaGraph2Vec learning strategy to learn latent embeddings of multiple types of nodes.

2

BEST FOR You�ORGANICS COMPANY

Dataset

Some popular datasets for HIN:

• Acm:
• ~ 7M words.
• Total embeddings: 11246
• yelp:
• ~ 14M words.
• Total embeddings: 37791

3

BEST FOR You�ORGANICS COMPANY

Algorithm

4

BEST FOR You�ORGANICS COMPANY

Generate heterogeneous neighborhood

Random walk

• It firstly counts the number of edge types satisfying the constraints and randomly selects one qualified edge type.
• Then it randomly walks across one edge of the selected edge type to the next node.
• If there are no qualified edge types, the random walk would terminate.

5

BEST FOR You�ORGANICS COMPANY

MetaGraph2Vec Embedding Learning

Skipgram

• To learn node embeddings, the MetaGraph2Vec algorithm first generates a set of metagraph guided random walks
• Then counts the occurrence frequency F(vi , vj ) of each node context pair (vi , vj ) within w window size.
• After that, stochastic gradient descent is used to learn the parameters

6

BEST FOR You�ORGANICS COMPANY

Training

Training for each epoch

• SGD optimization.
• Training with sample batches.
• Negative sampling.

7

BEST FOR You�ORGANICS COMPANY

Negative sampling

To speed up training, negative sampling is used to approximate the objective function.

8

BEST FOR You�ORGANICS COMPANY

Thank You

BEST FOR You�ORGANICS COMPANY