- Homogeneous networks: representative of singular type of nodes and relationships
- Challenges: multiple types of nodes and links
- Matapath2vec
- meta-path based random walks
- Heterogeneous skip-gram
- Matapath2vec++
- Structural and semantic correlations in heterogeneous networks.
Although there are different types of nodes in V, their representations are mapped into the same latent space.
Homogeneous network embedding
- Structural context = local neighborhoods
- Maximize the network probability in terms of local structures:
Heterogeneous network embedding: metapath2vec
- Heterogeneous skip-gram (model the structural correlations between nodes in a path)
- Meta-path-based random walks (Transform the structure of a network into skip-gram)
- A meta-path scheme
- Composite relations between node types
- Use meta-paths to guide heterogeneous random walkers, transition probability at step i:
-
- The flow of the walker is conditioned on the pre-defined meta-path scheme.
- The meta-path-based random walk strategy ensures that the semantic relationships between different types of nodes can be properly incorporated into skip-gram.
- Metapath2vec++
- Metapath2vec ignores the node type information in softmax. In other words, metapath2vec actually encourages all types of negative samples, including nodes of the same type t as well as the other types in the heterogeneous network.
- Heterogeneous negative sampling
-
- In metapath2vec++'s skip-gram, the multinomial distribution dimension for type t nodes is determined by the number of t-type nodes.
Relevance
- Word2vec
- Word2vec based network representation learning frameworks (homogeneous networks)
- DeepWalk
- LINE
- Node2vec
- PTE
- Negative sampling
- K-means algorithm
- Logistic regression classifier
- Biased random walkers (a mixture of breadth-first and width-first search procedures )