How Might We Automate the Generation of How-Might-We Questions: Computational Creativity in the Design Thinking Process
Open Access
Author:
Meinzer, Emmett
Area of Honors:
Elective Area of Honors - Engineering Design
Degree:
Bachelor of Science
Document Type:
Thesis
Thesis Supervisors:
Chris Mc Comb, Thesis Supervisor Sven G Bilen, Thesis Honors Advisor
Keywords:
natural language processing design thinking text generation
Abstract:
The detection and articulation of user needs can be a time-consuming and arduous task for designers. While social media offers a potential wealth of data from which user needs can be extracted, few data mining methods integrate seamlessly with standard design thinking approaches. This research establishes a methodology to assist designers in conceiving unique and viable solutions to solve design problems by automating the generation of ‘How might we…?’ (HMW) questions. The method uses a neural network composed of gated recurrent units to creatively generate HMW questions based on social media data. We also present a novel language discriminator using part of speech and token Markov chains to discriminate between useful and nonsensical HMW questions after each round of text generation. The results highlight the difficulty of decoding the complexity of human experience based solely on textual data.