Extracting Key Terms From Noisy and Multi-theme Documents.pdfVIP

Extracting Key Terms From Noisy and Multi-theme Documents Maria Grineva Maxim Grinev Dmitry Lizorkin Institute for System Institute for System Institute for System Programming Programming Programming of the Russian Academy of of the Russian Academy of of the Russian Academy of Sciences Sciences Sciences maria.grineva@ maxim@ lizorkin@ispras.ru ABSTRACT give high-level description of its content for readers. Key We present a novel method for key term extraction from text term extraction is a basic step for various tasks of natural documents. In our method, document is modeled as a graph language processing, such as document classification, doc- of semantic relationships between terms of that document. ument clustering, text summarization and inferring a more We exploit the following remarkable feature of the graph: general topic of a text document [10]. the terms related to the main topics of the document tend to Key term extraction is the core task of Internet content- bunch up into densely interconnected subgraphs or commu- based advertising systems, such as Google’s AdSense and nities, while non-important terms fall into weakly intercon- Yahoo! Contextual Match. They find relevant key terms on nected communities, or even become is