Information Retrieval System for News Articles in Russian

Abstract

We present a description of the natural language processing system developed for information retrieval project scan.interfax.ru. The system should process news articles in Russian and retrieve as much information as possible about persons, organizations or other text objects mentioned there. The conceptual system idea is to process and “understand” documents without using any time dependent named object databases. The system should retrieve and interpret person names, organizations, geography and some other text structures. And only on the second stage — to make identification of retrieved named objects via database. The resulting information is used in Scan project to allow complex-query search of news articles, to fill a named object database in automatic manner and to provide information for analytical services. We discuss functionality of the system, main approach ideas used and challenges to be resolved in future work. The article mostly covers entity extraction procedure and fact extraction mechanics. The problems of geography entity extraction are discussed particularly.