Data-driven and machine learning based design creativity

Abstract

The power of “big data” and artificial intelligence has advanced not only computer science but also other research fields. In this thesis, patterns, novel insights and knowledge of design creativity are explored and uncovered by exploiting huge, versatile and highly contextualized design data and advanced machine learning algorithms. Bisociation is applied to creative knowledge discovery along with network-based data mining and visualization techniques for exploring useful relationships and patterns between cross-domain concepts. In order to evaluate the proposed model, a web tool called B-Link has been developed in a longitudinal case study which shows its capability of augmenting creativity in idea generation tasks. In addition to the study of semantic creativity, a visual conceptual blending model is also developed for blending two semantically distinct concepts into image data, taking advantage of generative adversarial networks. This model is implemented in a design case study demonstrating its capability in generating images of a synthesized spoon and leaf for creative design. Taking combinational creativity as an example of design creativity, a novel approach for interpreting design creativity is introduced, in which image recognition and natural language processing technologies are investigated for key information extraction (e.g. combination pairs). A framework of reusing creative knowledge in a design creativity system is proposed, in which the functionality and relations of each module are fully illustrated. By integrating data, algorithms and creativity theories systematically, the framework shows the potential for recycling creative knowledge in a creative system for design.