0

I am interested in two things:

  1. First, how do I encode a subgraph from a knowledge graph as an input in a neural network, such that this is efficiently done and which type of neural network is efficient for this, if in the end I want to predict a real value from a certain subgraph?

The task should look like this: NN:

-Input: Knowledge Subgraph

-Output: Real value

How should the input specifically look? Is a neural network able to handle large data as this? Which neural network should I use?

  1. Secondly, since I know that there should be some subgraph embedding which will create small representations of subgraphs(not node embedding, but graph embedding), is there some useful approach which is able to embed Knowledge Subgraphs such that they can be provided as an input in NN?

Solutions for other similar directed graphs with different types of relations are also welcomed.

anascmidt
  • 1
  • not sure what kind of answer you expect now here, this question is clearly to broad and still research topic. There are plenty of approaches likegraph embeddings and GNN - feel free to read the papers about those – UninformedUser Jul 10 '20 at 05:52
  • @UninformedUser I am new at this topic and I don't know where to start. I have read several Knowledge graph embedding techniques but they provide Node embedding and not graph embedding. Also related to the neural network I just need an advice not research topic related. – anascmidt Jul 10 '20 at 08:08
  • That is not true. they also provide proper graph embeddings. There are *Sets of node embeddings and convolutional approaches* like Sum-Graph, Graph Coarsening or GNN. There are also existing tools, like [DGL](https://docs.dgl.ai/en/0.4.x/tutorials/models/5_giant_graph/2_giant.html) or something else. – UninformedUser Jul 10 '20 at 10:56
  • @UninformedUser this is what I actually need a suggestion for such approaches if you have some approaches in mind for Knowledge Graphs please provide them as an answer. Thank you. – anascmidt Jul 10 '20 at 12:24

0 Answers0