Brief introduction and advantages and disadvantages of deepwalk model

1、DeepWalk

[DeepWalk] DeepWalk- Online Learning of Social Representations (SBU 2014)

word2vec It's based on sequence embedding; however , In fact, the relationship between entities is becoming more and more complicated 、 Networking . This is the time sequence embedding------>graph embedding.

Definition of graph ：

G=(V,E),Evv

​

Two big pieces ：

1、 Build sequence ;-------> Random walk , Cut off random walk

2、word2vec Training ;------>Skip-gram

2. Model features ：

The article creatively puts forward random walk + skipGram The form of the node to express , A large number of industrial circles in the following years are concerned with emb The application of is based on this form .

deepWalk The core idea is that there are two steps ：
（1） Construct random walk sequence through relation network graph ;
（2） Pass the generated sequence through skip-Gram The formal memory of means , Each node can be represented in the same space . In this paper, the evaluation of algorithm effect is mainly through the task of multi label classification for users of several websites .
Experimental results show that , The improvement effect is obvious , At the same time, in the case of less training data, the algorithm can also have better performance .

advantage ：

When the data volume is sparse, it can still perform well

Support large-scale online execution of forecasts

Can realize parallel operation

The algorithms and formulas applied are based on the language model .

This paper proposes that if the training sequence can be obtained directly , It doesn't have to be a random walk .

3. reference ：

simrank:https://www.cnblogs.com/pinard/p/6362647.html
https://zhuanlan.zhihu.com/p/45167021

https://www.cnblogs.com/pinard/p/7243513.html
Original thesis ：http://www.perozzi.net/publications/14_kdd_deepwalk.pdf
Thesis translation ：https://www.jianshu.com/p/5adcc3d94159
Applied references ：https://yq.aliyun.com/articles/716011