Yandex launched the Archive Search service, which contains over 2.5 million pages of historical handwritten documents with text transcripts. This was reported by the company’s press office.
Handwritten pre-revolutionary documents are read by a neural network. It should be noted that experts trained him on a data board consisting of hundreds of thousands of handwritten lines of real texts from the 18th-19th centuries. from the Moscow Central Archives, as well as tens of millions of generated examples. Previously, they marked and deciphered historical materials themselves.
With the help of technology, users can find the desired document in the service’s database with the mention of a name, city, etc. Text search is available in the directory or through the search bar, there are also filters by years, archives, funds and stocks. The document decoded by the neural network is displayed to the right of the original, after hovering over the desired fragment, it will be immediately highlighted on the digital copy of the text, the press service explained.
In addition to the documents of the Moscow Central Archives, the database also includes texts from the archives of the Orenburg and Novgorod regions. The creators promise that in the future, even more documents will be available on the site.
In November 2022, Yandex also trained a neural network to find places for complex queries in Yandex.Maps. When processing a request, it analyzes not only the description and keywords, but also other parameters — photos, reviews and menus.
.
