Tuesday, August 27, 2013

More Puzzling Over Ancestry.com "Suggested Records"

After I wrote Puzzling Over How Ancestry.com Finds Suggested Records on 26 August, there were several comments from readers who thought that the "Suggested Records" were generated from records attached to someone's Ancestry Member Trees.  Some of the comments:

*  Sara Gredler said:  "Hmmm, my impression (in my experience and reading your workflow above) was that those Suggested Records *are* based on what has been attached to Ancestry.com trees, which is why on the 1880 census record showed up as a suggested record on other census hits but not others. I have always taken the 'Suggested Records' as a collective."

*  Geolover said:  "I agree with Sara's impression. Sometimes I see items for completely wrong names that would never be retrieved in a search. Since there is no link directly to others' savings, this is hard to check."

*  Taco Gouloose said:  "As far as I can tell, it's some algorithm that displays records that are attached to one individual within other Ancestry Member Trees, together with the record that you have currently selected. If a large selection of AMT's have the same set of records attached to an individual, Ancestry can offer up that set as 'suggested records'. The reason why it will show you wrong records sometimes, is the rampant, unverified copying of information from other AMT's by other Ancestry users, including wrongly sourced individuals. When enough people copy that wrong information to their own tree, it will show up as a suggested record at some point or other. Just my two cents."

I don't know if these folks are correct or not, and I went another step to determine if they were.

I searched for Ancestry Member Trees with George S. Knapp in them, using the same search I did yesterday:

There were only 3 Ancestry Member Trees with this George Knapp - 2 are Private and 1 is Public.

The two Private trees are:

One of them has no attached records, and the second one has 7 attached records.  I don't know which 7 records those attachments are, but they are probably U.S. census records.

The Public Member Tree is my own:

I have 2 sources, but no attached records in that tree.  

As I recall, I found 12 different records for this George Knapp using the search results and the "Suggested Records" hints (8 U.S. census records, and 4 Iowa state census records).  

Russ Worthington also has a Private Member Tree that he added when he helped me in late 2011 try to sort out the Knapp ancestry.  He attached the 1850, 1860 and 1870 U.S. census records to that tree.  However, this tree does not show up in the Private Member Tree list, probably because he used a setting to not search this tree.

The point is that my search results and the "suggested records" provide 12 found records, while the searchable Private and Public Trees provide only 7 attached records.  Where did the other "suggested records" come from?  I think that Ancestry found them by searching using the given search parameters.  

I don't know yet if Ancestry used the Ancestry Member Trees attachments - it's possible, and it makes sense that they would do that (heck, I would do that if I was programming this  (it's "low hanging fruit").  But then, since there were quite a few George Knapp's in the census records born in the 1846-1850 time frame (but only one in new Jersey), why don't those "not-the-right-George" records show up in the "Suggested Records?"  Several of them do, when the search engine found records that might be the right George Knapp.

If Ancestry.com is using Attached Member Tree attachments, then I wonder why the full list of "Suggested Records" isn't on every record summary.  In other words, for the 1880 U.S. census record, there were only 4 "Suggested Records" shown, so why aren't the other 7 "Suggested Records" shown (assuming they are all attached to an Ancestry Member Tree)?  

What if I was able to find someone in the records, that had "suggested records," but wasn't in an Ancestry Member Tree?  Would that be convincing that Ancestry.com is only doing a search and not using Ancestry Member Trees for "suggested records?"  Turn it around - what if I was able to find someone in an Ancestry Member Tree with attached records, but had no suggested records when I find a record for that person?

My hypothesis is that Ancestry.com is using search results AND attached records to Ancestry Member Trees for a specific person.  I think I've shown that they are doing searches.  I am not convinced that they are using Ancestry Member Tree attachments to drive "Suggested Records."  I think it's possible that they are not using any Ancestry Member Tree attachments as "Suggested Records."

I'm not sure that it matters a whole lot, other than for our wonderment and amusement, but it's something I've been thinking about.

The URL for this post is:  http://www.geneamusings.com/2013/08/more-puzzling-over-ancestrycom.html

Copyright (c) 2013, Randall J. Seaver


Geolover said...

If your theory is correct, why do you think there was such variability in the suggested records as shown in your first post? Why did not the search-engine results appear in each list?

I think it is possible that the basic mix is according to the search result you were theoretically considering attaching: the suggested records were compiled from those trees that had saved the contemplated-item. Some of the records could have been saved to trees that would not be retrieved by a search for "George S. Knapp." They could have been saved to a tree under just-George, under Stephen G., under Sidney G., etc.

Oh and the 7 items in the private tree possibly included Ancestry World Tree, Millennium File, and and other junk in addition to maybe census items (not necessarily correct ones), a vital record or two, findagrave, billiongraves, etc.

Kay Haden said...

I agree that Ancestry is using records attached to other trees. They certainly use the photos and stories that are attached to other trees that share your "person" and suggest them as records. Unfortunately the photos/stories do not always apply to that particular individual. I also get the suggested records for censuses, and war records, etc. that are obviously NOT that person - and there are too many ways this record doesn't match (even the name sometimes) for this suggested record to have appeared unless it was wrongly attached to someone, somewhere.

bgwiehle said...

Definitely data from member trees is used to generate the suggested records lists. Sometimes the list includes both maiden and married names of the person without any marriage record available in the historical records.
Something to beware of: using a chain of suggested records, i.e. your person > suggested record > suggested record > ... These quickly lose all connection to the original search parameters. Censuses aren't too bad (probably because other family members are used as matching criteria) but some record types (esp. directories) seem to look at just the name and approximate birthdate.

Donna said...

I only use the hints when I'm working on a specific person and only after I have searched for everything I found in searching. When I then come across a hint that, based on my research, is clearly for a different person, I try to track down the person(s) it is attached to - I have always found it attached to a person of the same or similar name I am researching. It has always been an attachment of one or two records that failed to give a complete picture of the person's family relationships and migration patterns.

Justin said...

Suggested records are definitely derived from a combination of both AMT and an additional strict search. I have seen suggested records for people that didn't exist in any AMTs, so they can't be purely generated from AMTs. I've also seen suggestions that were clearly generated from AMTs, as everyone else has seen too.