Use graph embedding

Whether or Network Graph embedding embedding is via node (node) and FIG side, embedding a learning vector for each node.

More popular algorithms:

Model	Paper	Note
DeepWalk	[KDD 2014]DeepWalk: Online Learning of Social Representations	[Graph Embedding] DeepWalk: algorithm theory, implementation and applications
LINE	[WWW 2015]LINE: Large-scale Information Network Embedding	[Graph Embedding] LINE: algorithm theory, implementation and applications
Node2Vec	[KDD 2016]node2vec: Scalable Feature Learning for Networks	[Graph Embedding] Node2Vec: algorithm theory, implementation and applications
SDNE	[KDD 2016]Structural Deep Network Embedding	[Graph Embedding] SDNE: algorithm theory, implementation and applications
Struc2Vec	[KDD 2017]struc2vec: Learning Node Representations from Structural Identity	[Graph Embedding] Struc2Vec: algorithm theory, implementation and applications

 

General application framework as follows:

1, FIG. Construction: Item item and the co-occurrence similarity matrix may constitute a network, wherein each item is a node, two item similarity reaches a predetermined threshold value has a direct connection side, of the degree of similarity smaller than the threshold is not connected.

2, walk strategy: inside the network, have come from a random node to the next node connection. Go several steps, we get a sequence of nodes.

deepwalk probability for selecting the next all edges connected to the node similarity value taken softmax, i.e. the probability scale value of 0 to 1. node2vec method combines DFS and BFS manner.

3, learning embedding vector:  the sequence into which random walk model learning word2vec give embedding vector for each node.

4, learned embedding vector machine learning methods for classification.

 

Links summary: https://github.com/shenweichen/GraphEmbedding