Multilingual Single Document Keyword Extraction for Information Retrieval

Abstract

Keywords play an important role in many aspects of information retrieval (IR). From Web searches to text summarization good keywords are a necessity. In a typical IR system algorithms are used which require the entire document collection to be built beforehand. While some research has been done on extracting keywords from a single document, the quality of the keywords was not based on how well they perform in IR tasks. Moreover, they are designed for only one language and the applicability to other languages is unknown. As such, this paper proposes a new algorithm that is applicable to multiple languages and extracts effective keywords that, to a high degree, uniquely identify a document. It needs only a single document to extract keywords and does not rely on machine learning methods. It was tested on a Japanese-English bilingual corpus and a portion of the Reuter's corpus using a keyword search algorithm. The results show that the extracted keywords do a good job at uniquely identifying the documents.