I am looking for a commendation on a hosted machine-learning process that provides an easy human-in-the-loop solution.
Our primary goal is to match strings from transactional data against our SSOT (Single Source Of Truth). Some of the matches are solely dependent on text/vector similarity (from transaction to SSOT) while others can incorporate multiple variables (State, County, ID) if the text/vector fails to match a string from the transaction to the SSOT.
Text/Vector case
From To Similarity
0 STRING STRING 1.0
1 STRING STRING 0.95
2 STRING STRING 0.85
3 STRING STRING 0.78
4 STRING STRING 0.70
State, County, ID case
From To Similarity
0 STRING STRING+State, County, ID 1.0
1 STRING STRING+State, County, ID 0.95
2 STRING STRING+State, County, ID 0.85
3 STRING STRING+State, County, ID 0.78
4 STRING STRING+State, County, ID 0.70
In this case, if the string match is poor in the Text/Vector case (below our threshold/human verification), State, County, and ID could help improve the match if we find an associated State, County, or ID to the string in question.
My web search yielded these generic answers:
AWS glue
Informatica PowerCenter
SQL Server Integration Services (SSIS)
Fivetran
Denodo Platform
FME
Alteryx Designer
Pentaho Data Integration (PDI)
Oracle GoldenGate
Someone also recommended AWS Sagemaker (and this is the best result so far)
I was expecting comparative results specifically related to string matching using machine learning. These results were useful for knowing what is out there but fail to give feedback as to the concrete experience on the matter, which is why I need a professional's guidance.
I am comfortable with Python3 and machine learning with sklearn.
