Crowd-sourcing and crowd-funding solutions like Threadless, Innocentive, Kaggle, and Kickstarter are transforming the creative design work, technical consulting, and venture capital. However, our empirical understanding of the processes of creativity, innovation, and decision making within these emerging socio-technical systems is sparse. Previous approaches to analyzing electronic brainstorming and group creativity have examined single-shot and ad-hoc small groups but not a community of thousands of users engaged in on-going idea generation and evaluation sessions. Moreover, prevailing approaches to assessing creativity rely upon abstract scenarios or the opinions of experts rather than large-scale voting or performance in a competitive market.
Our project involves developing a system to mine large databases of behavioral data from online crowdsourcing communities to make recommendations for users to remain engaged and predict the sales performance of users’ submissions. SkinnyCorp is a t-shirt retailer which operates a website called “Threadless” which crowdsources the creative design of t-shirts to members of its online community. Users propose graphical designs, receive feedback and comments from other users in online forums, submit revised designs for voting by the community, and SkinnyCorp, which owns the project, selects highly-voted shirts for printing and sales. Dozens of t-shirt designs are submitted every week which receive hundreds of comments and votes from other users and thousands of shirts and other products have been sold using Threadless’s “crowdsourcing” system.
Using this anonymized data involving hundreds of thousands of users repeatedly voting on creative designs submitted to an online community as well as the purchasing of products derived from these designs provides a profound test bed on which to evaluate existing approaches to collective creativity and decision making. These interactions among users, digital artifacts, and physical shirts can be structured as a complex multidimensional network involving user to user interactions (e.g., “who votes on whose designs?”), user to artifact interactions (e.g., “who comments on which popular designs?”), and artifact to artifact interactions (e.g., “what shirts are purchased together?”). Furthermore, findings gleaned from these analyses could inform the development of new methods and algorithms in network science for identifying factors which predict the success of creative work, unconvering how users become influential, and the processes governing the diffusion of innovations within a large community.