rahulnyk/knowledge_graph

Feature Extraction

Opened this issue · 2 comments

Hi,

Is there any way to extract embeddings for a part of graph?

For example we have Abstract India’s health indicators have improv... and we extracted knowledge like india's health indicators -> continue to lag behind -> peer nations. I want to extract some features for that knowledge.

There is a tool named Node2Vec, but it seems an old tool. I couldn't find any recent method.

Interesting Idea. I can only think of achieving this by embedding each triplet Node -Relation-> Node as text, into a vector. Would love to hear if you have a better idea.

In this work, they use GLOVE features. They extract features for tail nodes and edges.

Screenshot from 2024-05-01 16-51-26

I will integrate this idea into my work, but thinking of extracting GLOVE embeddings of head&tail&edge and pooling them.

As you can say, sentence embeddings can also be used