Conceptual Clustering of Patents

Abstract

Within technological R&D companies, the decisions made within their Strategic Technology Planning (STP) process are of great importance due to their financial, legal and competitive repercussions. For this reason their R&D managers want to be as widely informed as possible about the risks and benefits of developing a certain technology. As the subjects for which these decisions are made, are too complex to comprehend, there is exploitation of so-called Decision Support Systems. This research focusses on those Decision Support Systems in the STP process which are based on patent analysis. Patent analysis is relevant to apply in this process as the patent database is considered to be a comprehensive collection of technological concepts. This is due to the fact that companies voluntarily deliver the information, and benefit from describing their technological innovation as complete as possible. Throughout the STP process multiple needs for insight to make informed decisions can be detected. However, as every applied form of analysis will present a representation of the selected subject, which’ outcome is based on assumptions made as well as on the underlying method applied. The user cannot be sure that this given outcome is always correctly representing reality. Hence, they are applying multiple representation methods on the same subject, a process which is also referred to as multiplism. Within this process they are searching for robust outcomes that hold true under all the different representations. Due to this process, there is an ever present interest in new methods of representation and analysis, to extend the R&D division’s decision support toolkit. Due to the ever present need for more insight, alongside a methodological interest, this research focusses on the evaluation of the applicability of conceptual clustering to patents, as well as its potential usability in the STP process. Conceptual clustering is a machine learning technique developed in the 1980’s, which automatically categorizes the entered information, resulting in a tree shaped hierarchy. The conceptual clustering technique hasn’t been found to be applied to patents before and this research will therefore also focus on identifying its potential use within the STP process. To gain insight into the different needs for patent analysis within the STP process, as well as to provide a source for evaluation, a design has been made for the decision support system by applying the Axiomatic Design method. The inputs for this design are customer needs which have been based on a literature study and expert input on the different applications for patent analysis in the STP process. To assess its potential applicability, a prototype Decision Support System, constructed around the method of conceptual clustering applied to patents, has been developed. For which the implementation of the conceptual clustering functionality is done based on the COBWEB algorithm as proposed by Fisher. The prototype entails two software products written in Python which are respectively a text-mining and a conceptual clustering program, combined with a visualization tool based on the D3 JavaScript library, for which an online environment has been developed. To illustrate the evaluation, a complete selection of patents related to additive manufacturing (3D printing) retrieved from Thomson Reuters’ Derwent patent database is used. This patent set (containing 9360 patents) has been parsed through-, and visualized by-, the developed prototype, showing a representation of this technological field’s inter-patent relational structure. Through evaluation based on the design, the prototype has been proven to be usable within the STP process for Exploration analysis, Competitor analysis and Portfolio analysis. The highest added value to the R&D division’s toolkit is perceived to lay in the explainability of the outcome, the accessibility through the online visual representation and the possibility of using the decision support system in an interactive way. Allowing the technology scouts and managers to interact with the data, discussing and continuing exploring based on newly gathered insights. Furthermore implementable solutions are presented, allowing to extend the developed prototype’s applicability for use to the inclusion of Freedom to operate analysis capabilities, trend analysis and to a limited extend inventive problem solving analysis capabilities. Further evaluation of the prototype based on a comparative analysis of the case with the output created with a self-organizing map technique, leads to further conclusions on the conceptual clustering technique. Firstly, the same high level clusters can be detected, however not all inter-relations of these clusters match. Forcing the user to further investigate, increasing robustness of knowledge obtained. Secondly, the conceptual clustering shows more application domains in addition to the technical concepts. Thirdly, multiple representations can be made of the same set based on user needs. Fourthly, the influence of the attribute number, for which it is suggested to perform a high level search first. Then select a subset of patents for reprocessing and visualization, in order to obtain more insight into sub-branching. Finally, the combinational use of the two methods enhances the insight, due to the explanatory value of the nodes in the conceptual clustering output.