Enhancing financial crime and compliance with ML for entity resolution | CloudWorld 2022

Find out more: oracle.com/ind...
We present a few ways to enable the compliance objectives in Oracle Financial Services Crime and Compliance Studio. First, we show how our enhanced entity resolution employs sophisticated machine learning techniques to analyze and precisely match the names and addresses of different individuals or organizations across different data sources. Second, we talk about how to use graph machine learning to detect similar investigation cases (represented as subgraphs) to help investigators easily correlate new cases with historical ones.
Watch Marcel Gwerder of Oracle discuss enhancing financial crime and compliance with machine learning for entity resolution at CloudWorld 2022.
Contact Oracle Sales: social.ora.cl/...
Oracle: social.ora.cl/6...
Try Oracle Cloud for free: social.ora.cl/...
Oracle Events: social.ora.cl/...
Oracle Support: social.ora.cl/...
Oracle's communities: social.ora.cl/...
Subscribe to Oracle's UA-cam channel: social.ora.cl/...

[@mwredfern]

You can do the similarity matching without any ML functions. Oracle itself has a similarity function you can use in the where clause, giving you a percentage match.
And although this kind of matching seems like a useful means to location matching, you’ll find that the match level will eventually be extremely low.
If you ever have resort to location matching, just know that you’re dealing to bad data in, bad data out.
If I was working this issue and came to point where I only had location match as my next step, I would stop here and contact the DBA and ask to see the source tables and schemas for the data you are using. There’s a chance that the source data has additional data you can use to accomplish your goal of matching. It’s a small chance. But it’s there.
Bottom line, string and/or fuzzy matching rarely helps in getting the ball moving forward. This a bad data issue. You might pick up a few matches here and there. But when you try to sell this to the VP’s, they’re always gonna ask for primary key/identifier matching as proof. Cause no one makes decisions on a hunch. 😊