关键词: 背景知识, 关键词抽取, 专利数据, 支持向量机, 信息检索

Abstract: Keywords are important clues that can help a user quickly decide whether to skip, to scan, or to read the article. Keyword extraction plays an increasingly crucial role in information retrieval, natural language processing and other several text related researches. This paper addresses the problem of automatic keyword extraction and designs a novel automatic keyword extraction approach making use of patent knowledge. This approach can help computer to learn and understand the document as human being according to its background knowledge, finally pick out keywords automatically. The patent data set is chosen as external knowledge repository because of its huge amount of data, rich content, accurate expression and professional authority. This paper uses patent data set as the external knowledge repository serves for keyword extraction. An algorithm is designed to construct the background knowledge repository based on patent data set, also a method for automatic keyword extraction with novel word features is provided. This paper discusses the characters of patent data, mines the relation between different patent files to construct background knowledge repository for target document, and finally achieves keyword extraction. The related patent files of target document are used to construct background knowledge repository. The information of patent inventors, assignees, citations and classification are used to mining the hidden knowledge and relationship between different patent files. And the related knowledge is imported to extend the background knowledge repository. Novel word features are derived according to the different background knowledge supplied by patent data. The word features reflecting the document’s background knowledge offer valuable indications on individual words’ importance in the target document. The keyword extraction problem can then be regarded as a classification problem and the support vector machine (SVM) is used to extract the keywords. Experiments have been done using patent data set and open data set. Experimental results have proved that using these novel word features, the novel approach can achieve superior performance in keyword extraction to other state-of-the-art approaches.

Key words: background knowledge, keyword extraction, patent data, support vector machine (SVM), information retrieval