What the Structure of Crowdsourced Solution Landscapes Reveals About Their Ideas’ Value
Julian Just, Thomas Ströhle, Johann Fueller, Katja Hutter · Academy of Management Proceedings · 2024
In crowdsourcing contests, participants span rich solution landscapes where ideas take different locations depending on their semantic similarity. The characteristics of an idea derived from its location in the solution landscape, e.g., its semantic distinctiveness or network centrality, can affect its likelihood of success. Despite increasing research efforts using different measurement approaches and contextual variations, we lack a thorough understanding of the potentially complex relationship between idea similarity and value. This study comprehensively examines how the similarity structures of ideas embedded in solution landscapes influence the likelihood of success. Using a dataset of 11,615 ideas from 39 crowdsourcing challenges, the research combines natural language processing (NLP) to retrieve multiple semantic-similarity-based idea features and machine learning algorithms to inductively derive patterns and validate them in a hold-out sample. The results show that while distinctive ideas are generally more likely to be successful, associations are non-linear and context-dependent. The study highlights the importance of the density of the spanned landscape. In denser landscapes, distinctiveness in terms of word atypicality is negatively related to success. While highly interconnected ideas are more appreciated, ideas closely connected to others are less successful in dense contexts. By providing a more nuanced view of the role of similarity structures in idea success, the research advances the theoretical understanding of search across crowdsourced solution landscapes. As such, it contributes to a more strategic approach to idea generation and evaluation in crowdsourcing environments.