Robert Kehrer of FamilySearch read that blog post, and provided additional information about the Record Hints feature via email:
"I noticed that you will be evaluating the quality of the hints in an upcoming article. I thought I’d give you a little insight to enrich that article.
- Hints are generated by taking all the information (name, gender, event dates and places, relationships) from the ancestor and that ancestor’s one-hop-relatives (parents, spouse, children)
- The Family Tree persons are matched against all 4.5 billion historical record personas to identify the highest confidence matches
- Hints are being exposed incrementally.
- Today we deliver only hints where the matched person is primary on the record.
- We will shortly expose hints to secondary personas
- New records, improved algorithms, etc will ensure there are always new hints to evaluate.
- The message to users is “Don’t be surprised if you find a record through searching that wasn’t exposed as a hint. Keep coming back to the site to see new hints”.
- Part of the reason for this is the current release is a “Public Preview”. The hinting system will still be a work in progress through late in the year. We will be working on a number of key things:
- Continuous improvement in algorithms
- Update system systems to automatically re-hint changed tree people and newly published records (the current system uses snapshot data updated monthly)
- Optimizations for hinting diverse language collections
- Because it uses snapshot data, an astute researcher may notice a couple things:
- A newly added person to the Family Tree may not get new hints for as much as a month
- A significantly changed Family Tree person may display hints that no longer perfectly match the tree person data for as long as a month
- As a measure of accuracy, we have been watching the rate at which users mark presented hints as Not-A-Match. That rate has been exceedingly low.