Uggestion need to use stellargraph on Visual rich document like invoice

I’m very interested in using Stellargraph GNN on Visual rich documents like invoice, identity card, etc.

I’m trying to build invoice extraction as a hobby project using GCN. I have 250 invoices which have SellerName, SellerAddress, Total Amount and other things.

I have used OCR to extract the bounding box of every word from each invoice and build graph. Graph is built by seeing what word is side and below.

Now I have 250 graphs, but most of the nodes are labeled as “Others”, because i’m not interested in these nodes.

Approx. distribution of the labels looks like,

SellerName - 500

SellerAddress - 1500

Total Amount - 500

Others - 12000

Based on my understanding, there are two types of Graph; inductive and transductive.

Can you please suggest how to proceed with Stellargraph to build Information extraction using GNN.

Hi! Have you seen StellarGraph’s demos? Based on your description, the most relevant ones would probably be:

There’s probably various academic papers that are relevant for inspiration too:

We’re happy to help if you have more specific questions, or have some code you want some advice on! :smile: