April 12, 2007
Internet, April 12, 2007. From date, Yandex.News offers a new page for the news feature (selection of news related to one event). A combination of main facts, illustrations, and reference information creates three-dimensional representation of the news.
The facts are being selected from accounts of various sources, which allows to introduce different viewpoints. At that, the informative, current, and exclusive news items are privileged. The news feature is illustrated with photographs and maps (if some geographic objects are mentioned). The names of the key figures in the feature are directly linked to their press-portraits.
The new feature continued the development of clusterization algorithms, automatic summarization and fact extraction technologies used in Yandex.News. It is based on the technology which we called „multi-document summarization“. „From all reports in the feature, the most important objects, like personal names, company names, geographic objects, dates and numbers, are selected automatically. These, together with feature's key words and news queries, determine the choice of texts for summarization“, says Lev Gershenzon, leading developer of Yandex. „The factographic approach to summarization, as we see it, makes it more informative.“
Yandex.News service has contractual partnership with 1,700 information sources and it processes 100,000 reports daily. The main objective of Yandex.News is to automatically create a full information picture of the day. „The reader gets the notion of what has happened right from the feature page, and to receive detailed information they can choose the source they like better“, says Tatiana Platonova, head of Yandex.News service. „And the source, our partner-website, has an interested visitor.“
News processing and systematization are automated. Algorithms for data analysis and ranking is an original development of Yandex, which is based on identification of textual similarity and clusterization of information streams. Fact extraction technology was first used at Yandex.News for automatic selection of key people in the news reports. A year ago the public gained access to press-portraits, reference information about people, automatically extracted from mass media news. Automatic address and quotation selection and identification is also based on this technology.
