Mickey

Also I tried with my own training data along with zingg's automated training and labelling twice. I noticed two things: somehow spark ran very slowly. And the accuracy went really bad. Somehow it grouped some totally unrelated persons wrongly

But it does not mean everything will be present in a record always

Nope. It has many fields pii like name, phone, email, address etc

I tied zingg labeling but somehow it does not group these sample example together : N Johnathon, 999-999-9999, abcd@gmail.com Nathan Johathon, 777-777-7777, abcd@gmail.com Nathen Johnathon, 999-999-9999, xyz@gmail.com All three records belong to same group or person due to common determinstic attributes, like phone number and email id along with similar names. But zingg match them as two differt persons !

Thanks! I went through the link above. What I understand that, incremental logic is only available on paid version with zingg id. For opensource, we need to run the full load again where clusterid may change after regrouping. Is it the correct understanding?

Got it. Thanks! For incremental load, what happens to a record which belongs to an existing historical group id? Does it assign the old clusterid? How to make sure it uses the consistent for similarity but different clusterid for new, on every run?