3 ms·
Mainly OCR. We were looking for cases where a specific set of law required forms were missing. It took 14 days to train the algorithm to recognize these forms,
by eksemplar 10y ago
Mainly OCR. We were looking for cases where a specific set of law required forms were missing.
It took 14 days to train the algorithm to recognize these forms, and maybe 1,5 months to learn how to use the algorithm and the Azure cloud.
I can't give you the exact technical details because I'm not the CS guy, but I can tel you that we used Googles OCR and a common tree algorithm suggested by "the internet" for machine learning.
The machine vastly outperformed humans in most cases, especially with 50+ page compilations, but got beat when documents had been scanned extremely poorly.