Text Search is not finding all instances of a name
I am searching for the name "Edward Davidson" in Hancock County Kentucky in the years 1850-1900. In doing so, the search returned several different pages. On of those pages with the one listed in the URL https://www.familysearch.org/ark:/61903/3:1:3Q9M-C37Q-PWVL?view=fullText&keywords=Edward%20Davidson%2CKentucky%2CHancock&groupId=TH-909-85642-86314-97
This is a Deed from "Hancock. Deeds 1883-1885, 1882-1883". Although Edward's brother, Benjamin C Davidson is mentioned in this document and really is the focus of this document, it references Edward Davidson in another Deed book.
The last line highlights "Edward Davidson" and it continues "and wife in Deed Book 14"
continues on next page
"page 375".
The deed book is listed on the wiki for Hancock, County, Kentucky.
I went to this collection and then to page 375. https://www.familysearch.org/ark:/61903/3:1:3Q9M-C37Q-LS3W-8?cat=105102
This particular page clearly has "Edward Davidson" listed with his wife "Sarah Katherine" on the second line along with brother "Benjamin C David" - "son" with "son on next line.
This particular image was not picked up with the Text Search for "Edward Davidson". So I hope I am giving your engineers enough info here to recreate the issue.
Risposte
-
Mod note: your post was edited to correct link errors. For future reference, links given their own lines often get corrupted. This means https://www.familysearch.org/ark:/61903/3:1:3Q9M-C37Q-LS3W-8?cat=105102 shows up like this.
0 -
Using machine learning technology to read handwritten documents is an "imperfect science" and errors, misreadings, missed words, etc. are to be expected. Researchers will still need to get their "eyes" on the written text to verify the transcription and to find instances of error. Regardless, it is a vast improvement over manually reading millions of images.
2 -
Before retiring, I worked on a system which used form of AI called an expert system which identify fraudulent call patterns with one of the major Telcos. I worked there for 25 years as a lead. I'm retired now so hopefully as I find issues, I can document them in a fashion that allows your developers to identify problem areas. I understand what a fantastic tool this is and will be as you work out the bugs. To many people 'AI' is new but developers have been working in the area for years. So if there is anything additional information that would be helpful when I have an issue like this, please let me know. We took real time feeds of metadata from the telco switches so what you are doing is different.
1
