Automatics tools and methods for patents analysis: efficient methodology for patent document clustering

Abstract

Patents have become a potentially powerful data sources and a wealth of information for companies and organizations that tend to analyze and exploit them for a variety of purposes and interests. However, with the ever-increasing volume of patents filed year after year and the multiplicity of patent global data bases, the task of patent research and analysis is becoming increasingly complicated and traditional analytical approaches have shown their limitations and have become costly in terms of time and labor. Thus, various techniques and approaches have been proposed to help specialists in their tasks of patent collection, analysis and results visualization. Document Clustering is one of these common technics widely used in patent analysis. Over time, several powerful clustering techniques and algorithms have emerged, but they often require modifications and adaptations depending on the fields of application and the target data. In view of this, we propose in this paper a methodology for obtaining an efficient clustering for patent documents based on the k-means, k-means ++ algorithm and various data-mining and text-mining techniques. Commons issues often faced during the analysis of patents such as the manipulation and representation of textual data or the curse of dimension will also be addressed in this study.