Tian Chan, Jurgen Mihm, Manuel Sosa
This paper introduces a structured approach, based on clustering analysis, to identify groups of visually similar designs (and potentially members of a design style) in a large set of unique designs. We implemented our approach on the population of design patents filed in the USA during the 1976-2010 period. Design patent data is used in our analysis for two reasons: (1) claims of design patents concern the visual aspects of classes of artifacts; (2) similarity between design patents is captured through a tightly controlled citation process. The cluster analysis is based on constructing a similarity index across all pairs of design patents, followed by applying a divisive graph clustering algorithm to organize design patents into a hierarchical tree. Identified clusters of potentially visually similar design patents are tested for positive association against other grouping criteria that are expected to contain similar designs such as design patents filed by the same designer, design patents filed by the same firm, and design patents assigned to the same patent class category. We discuss the implications of our findings for the study of the emergence and evolution of styles in design. Copyright © 2012 by ASME.
Technology and Operations Mgmt, INSEAD, Singapore; Technology and Operations Mgmt, INSEAD, Fontainebleau, France
