Natural language processing versus rule-based text analysis: Comparing BERT score and readability indices to predict crowdfunding outcomes

We explore how natural language processing can be applied to predict crowdfunding outcomes. Using the Bidirectional Encoder Representations from Transformers (BERT) technique, we find that crowdfunding projects that use a story section description with a higher average BERT score (indicating a lower quality of writing) tend to raise more funding than those with lower average BERT scores. In contrast, risk descriptions that have higher BERT scores tend to receive less funding and attract fewer backers. These relationships remain consistent after controlling for various traditional readability indices, highlighting the potential benefits of incorporating natural language processing techniques in entrepreneurship research.