Machine learning for creativity
Wei, Y. “Max,” Hong, J., & Tellis, G. J. (2022). Machine Learning for Creativity: Using Similarity Networks to Design Better Crowdfunding Projects. Journal of Marketing, 86(2), 87–104. https://doi.org/10.1177/00222429211005481
First Pass
Category
What type of paper is this?
- Empirical research, testing the application of machine learning to assess combinatorial creativity in achieving success for the design of crowdfunded projects
Context
What other papers is this one related to?
- Built upon prior studies on crowdfunding and combinatorial creativity
Correctness
What are the assumptions and are they valid?
- Assumes similarity between crowdfunding projects is a proxy for combinatorial creativity can adequately predict success
- Assumes that success of a project can be measured via funding amount
Contributions
Who are the paper's main contributors?
Clarity
Second Pass
Methodology
- Project similarity measured by Machine Learning of words (Word2Vec) and semantic similarity of documents (Word Mover's Distance)
- Similarities between projects of different categories have larger computational costs as they have to account for cross-category similarities, but may improve predictions.
- Networks' general design
- Time complexity introduced into network by varying weight between the 2 nodes by the time gap between the 2 projects
Metrics:
- The role of novelty in new product performance
- Prior success rate
- Goal overshoot
- Atypicality
- It is difficult to define atypicality with a weighted network, but we can work with an unweighted network by defining an arbitrary cutoff and examining which pairs falls below this cutoff.
Findings
- A higher prior success residual (modified prior success metric) is a significant indicator of better funding outcomes. Success begets success
- Inverted U-shape outcome for effect of amount of prior similarity to success. A project benefits most from a balanced level of novelty. Ideas that are too novel or too formulaic hurts. Balance in everything
- Some studies suggest that novelty has a positive effect on success (e.g., Dahl and Moreau 2002; Sethi, Smith, and Park 2001), some suggest a negative effect (e.g., Hyytinen, Pajarinen, and Rouvinen 2015), and others suggest less clear relations (e.g., Im and Workman 2004).
- Neither too little nor too much atypicality benefits funding performance. Thus, the creator should keep a balance between conventional and atypical combinations of prior ideas.
- Paper includes some more insight on managerial recommendations