In NLP-based street address parsing, training new rules to enhance existing models is as much a dicey proposition as a necessity. The problem is that the additions of the newly-trained rules may unexpectedly nullify previously-trained-and-established rules in the model and may lead to a lower overall success rate. The presenter will share a training and evaluation system built with FME Server at its core supplemented with Google Data Studio for monitoring and assessment to maximize the NLP training performance and outcome.
Gistic Research, Inc.
FME User Conference 2022