June 23, 2006
Internet, June 23, 2006. Starting today, the stories in Yandex.News that mention geographical addresses related to events also feature relevant fragments of the city map. The retrieval of addresses in the news and selection of matching map fragments for illustration are based on a proprietary fact extraction technology developed by Yandex.
Addresses are usually given in regional news: examples can be a construction of a new building, a street performance, or a car accident. Events are associated with maps in the news from Moscow, Ekaterinburg, Kiev, and other regions. In total, addresses are processed in ten cities.
"There are times when you read the news and find out that something has happened or will occur in your city, and then you wreck your brain on where exactly that side street is. Now the user can see the exact place on the map," comments Tatiana Platonova, head of Yandex.News service. "For a closer look, a single click would take him or her to the larger map at Yandex.Maps."
The whole process, from finding and extracting the address in the news text to making an illustration to the story, is automatic. It employs an index of words denoting address objects (street, square, bridge, etc.). This approach helped create a multi-purpose tool that is independent of a particular city address database.
"From the point of view of technology, an interesting challenge lay in identifying the city of the address mentioned in a news item. The thing is that news writers seldom put the city name right next to the street address," says Lev Gershenzon, leading developer of Yandex company. "The solution to this problem required extracting names of all settlements and then selecting the most relevant one, depending on the number of instances in the document, location in the document, etc."
The fact extraction technology was first used in Yandex.News almost a year ago for automatic identification of key people covered in news items and creating automated ratings of the most popular personalities (“Names in Yandex.News”). In spring, Yandex provided press portraits — background information and references on people automatically retrieved from mass media news. Automatic address extraction and matching is built on the same technology.
Note: This news item mentions no addresses besides "the Internet". However, if for example it featured the Yandex office address in Moscow — 1 Samokatnaya St., it would display the following map fragment next to it:
