Work

AI & Innovation Management : comparing human and machine-based similarity representation of ideas and evaluating the granularity of solution landscapes in the Fuzzy Front End of Innovation

Year: 2021
Type: dissertation
Abstract: With the rapid advances in digitization, innovation no longer principally takes place via traditional research and development avenues; today it is mainly undertaken on crowdsourcing platforms. A result is that innovation managers face a vast number of ideas, and the processing and analyses of all this information in order to gain an overview has become tedious, costly, and labor-intensive.\n\nThe thesis investigates the role of machine learning in overcoming this information overflow and how it can assist information processing in the fuzzy front end of innovation. In this context, the study explores how humans and the state-of-the-art machine learning algorithm BERT (Bidirectional Encoder Representations from Transformers) summarize content and group ideas according to their perception of similarity. Finally, the thesis asks which level of pre-processed information (granularity level) in the form of clustering adds most value in the innovation evaluation process.\n\nThe research of this thesis follows a quantitative approach with a secondary data set of a crowdsourcing innovation contest from OpenIDEO. From an experimental setting the study observed human’s limited processing capabilities and the phenomenon of cognitive load when processing ideas of a crowdsourcing contest and compared them to output of transformer BERT.\n\nThe analysis shows overlaps between the “average” human and BERT when summarizing ideas with keywords. Further, results show that human similarity assessment is partially similar to machine-based similarity assessment. \nBy looking at the results in detail they reveal that while every human and machine-based approach is individual and unique, every machine-based clustering solution is useful, adds value and significantly accelerates the processing of ideas. Finally, results imply that pre-processed, clustered ideas add most value when the number of clusters itself is small and the number of ideas within the groups is large. Thus, according to the findings of this study, a high level of granularity is most useful. \nThe study concludes that today Machine Learning already possesses all the necessary tools to support and speed up the low value tasks of idea processing and evaluation in the era of open innovation.
Author: Frommelt, Constantin Paul Egmond
Language: English
Open Access status: green
OpenAlex ID: W7017249484
Cites: 141
Cited by:
Related to: 10