The need for templates

[courtenayparserr]

I havent played with invoice2data yet but i wondered whether you have to have templates?

Is there any way one could maintain a db of possible terms for each distinguishable entity eg. Invoice number = invoice no

In that way it could test all terms until it found a match?

[m3nu]

Don't think that this will work because you can't map the keywords to values without doing proper parsing of the text.

[m3nu]

There are some ways to use machine learning for matching. They produce around 70 to 80% accuracy right now. Probably not enough to use them in a business setting because everything will need to be rechecked. So investing in templates and getting 100% accuracy is probably better for now.

Here some literature if you are interested @courtenayparserr

• https://medium.com/tradeshift-engineering/scaling-up-machine-learning-algorithm-for-form-recognition-bd09b319e14a
• http://cs229.stanford.edu/proj2016/report/LiuWanZhang-UnstructuredDocumentRecognitionOnBusinessInvoice-report.pdf
• https://arxiv.org/pdf/1708.07403.pdf