Document Understanding ML training for 100 vendor

Help AI Center
,
1

I have +100 vendor invoice’s and I have 100 sample for each vendor. Now I am trying to extract the needed information from them using ML and while creating the dataset and labeling them and then train the model through the pipeline it takes a lot of time and effort and I think the maintains is very difficult. Can you imagine if one vendor change an field or something like that.

My question is : Should I make a different Dataset, Data Labeling, ML Package, Pipeline,etc… for each vendor? it looks much easier this way and I think by doing so, if I am having an issue with the accuracy with one vendor, I can focus on that and get a better results.

What do you think?

2

@ahmed.emad

You can use multiple as well if you feel that is easy…but again cost comes into picture more deployments

Now coming to one type …as you would anyways need to use classifier the over mode accuracy might come down but the model can still perform wel on docs ehich dint change

And coming to retraining etc…even if it is separate or same the efforts are only on the docs that changed and not on all in either case

Cheers