A new semantic and machine learning approach to patent valuation Dr. Bernd Drapp, Head of Innovation and IP Management, Fraunhofer-TEG

Abstract

The value of something cannot be stated in the abstract; all that can be stated is its value in a particular place, at a particular time, in particular circumstances (King, 2001). Following King's statement the value of a patent is not a fixed number. It is influenced by various factors and certainly by the motivation for the valuation step, which may differ from case to case. Eg. selling a patent or buying a patent or strategic decisions on R&D investments may cause a valuation. Due to the different motivations there is neither an agreed formula nor a common approach for the valuation process, especially when the strategic value is asked instead of the monetary value. However, the management of intellectual assets has become a central feature in the new knowledge-based economy and so-called patent value indicators were gathered and classified in the past. Most of these indicators are based on bibliographic- and patent meta data as the length of the independent claims or the number of IPCs, repectively.

In our paper we present a new approach for the classification and valuation of patents which combines machine learning techniques, text mining and the semantic analysis of a patent's content. Following our approach the value indicators are no longer limited to bibliographic or mata data. Indicators such as: "Kind of Patent (e.g., method, process or substance patent)", "Technical sophistication", "Width of the claims" or "Object of protection" can only be obtained by means of content distillery techniques. Text mining and the semantic analysis of a patent's content were combined for that purpose. The content related indicators were gathered from a variety of Fraunhofer patents with a known value. Test samples were generated and machine learning techniques were used for the valuation step of the patents. Different algorithms based on statistical methods as well as support vector machines were tested for valuation. Further investigations pointed out the influence of different types of value indicators to the valuation. We also report how a neuronal network, text mining tools and the semantic analysis of patent were combined to achieve a flexible and powerful patent valuation module.

Speaker Biography

Dr. Bernd Drapp joined Fraunhofer-Gesellschaft in 2005 after having worked at two large international technology corporations. He is heading the Innovation- and IP-Management at the Fraunhofer-IAO. The group carries out industrial projects in almost all IP related areas including IP-valuation and liquidations. The group's research interests address the usage of patent data along the innovation process and technology analysis by the means of patent data. New IT-based tools are developed and tested for that purpose.