We expect this new opt-in function to help improve match accuracy. The new function is opt-in, and we’d like to invite you to see if it works better for you. The new rule-based scoring mechanism is called logic-v1, and it has been significantly enhanced to provide more accurate matching.
As our matching system evolves, we’re becoming more and more focused on rule-based and explainable mechanisms that consider a broad range of inputs to generate strong evidence about the risk associated with a screened entity.
Today in technology blog post visual metaphors: are these butterflies the same?
To keep it simple: With our API, if you pass ?algorithm=best, you will always get the most precise scoring that the current version of yente can support, but the specific score numbers will change as we update the software. Not specifying the ?algorithm= parameter will use the first model we implemented, regression-v1, for compatibility reasons.
Try this out against our SaaS API, or on-premise by updating yente to the latest version and using syntax like this:
POST /match/sanctions?algorithm=logic-v1
The logic-v1 algorithm differs from our existing algorithms in the following ways:
logic-v1changes how we compare names with many different parts, and is much more conservative in what constitutes a match. This is particularly true of name comparisons where both names have a very different length.- We're switching to metaphone as an algorithm for phonetic comparison in
logic-v1. It provides more precise matching for non-English names. logic-v1supports identifier-only matching for vessels (IMO), securities (ISIN), some company identifiers (OGRN and INN) and crypto wallets (public wallet address).- Providing a separate
lastNamefor persons can now also be used to reduce false positives.
You can try out the system using out advanced search web
We’re looking forward to your feedback, preferably in the #technical Slack channel or via support@opensanctions.org. We’re eager to hear about your results.
