Ancestry Taps AI To Sift Through Millions of Obituaries

Algorithms identified death notices in old newspaper pages, then another set of algorithms pulled names and other key details into a searchable database. From a report: Ancestry used artificial intelligence to extract obituary details hidden in a half-billion digitized newspaper pages dating back to 1690, data invaluable for customers building their family trees. The family history and consumer-genomics company, based in Lehi, Utah, began the project in late 2017 and introduced the new functionality last month. Through its subsidiary Newspapers.com, the company had a trove of newspaper pages, including obituaries — but it said that manually finding and importing those death notices to Ancestry.com in a form that was usable for customers would likely have taken years. Instead, Ancestry tasked its 24-person data-science team with having technology pinpoint and make sense of the data. The team trained machine-learning algorithms to recognize obituary content in those 525 million newspaper pages. It then trained another set of algorithms to detect and index key facts from the obituaries, such as names of the deceased’s spouse and children, birth dates, birthplaces and more.

Ancestry, which has about 3.5 million subscribers, now offers about 262 million obituaries, up from roughly 40 million two years ago. Its database includes about a billion names associated with obituaries, including names of the deceased and their relatives. Besides analyzing the trove of old newspaper pages, the algorithms were also applied to online obituaries coming into Ancestry’s database, making them more searchable. Before the AI overhaul, the roughly 40 million obituaries on Ancestry.com were searchable only by the name of the deceased. That meant a search for “Mary R. Smith,” for instance, would yield obituaries only for people with that name — not other obituaries that mentioned that name as a sibling or child.

Source: Ancestry Taps AI To Sift Through Millions of Obituaries – Slashdot