Discovering Structure in Design Databases Through Functional and Surface Based Mapping

Abstract

This work presents a methodology for discovering structure in design repository databases, toward the ultimate goal of stimulating designers through design-by-analogy. Using a Bayesian model combined with latent semantic analysis (LSA) for discovering structural form in data, an exploration of inherent structural forms, based on the content and similarity of design data, is undertaken to gain useful insights into the nature of the design space. In this work, the approach is applied to uncover structure in the U.S. patent database. More specifically, the functional content and surface content of the patents are processed and mapped separately, yielding structures that have the potential to develop a better understanding of the functional and surface similarity of patents. Structures created with this methodology yield spaces of patents that are meaningfully arranged into labeled clusters, and labeled regions, based on their functional similarity or surface content similarity. Examples show that cross-domain associations and transfer of knowledge based on functional similarity can be extracted from the function based structures, and even from the surface content based structures as well. The comparison of different structural form types is shown to yield different insights into the arrangement of the space, the interrelationships between the patents, and the information within the patents that is attended to—enabling multiple representations of the same space to be easily accessible for design inspiration purposes. In addition, the placement of a design problem in the space effectively points to the most relevant cluster of patents in the space as an effective starting point of stimulation. These results provide a basis for automated discovery of cross-domain analogy, among other implications for creating a computational design stimulation tool.