The Predictive Power of Latent Variables: An In-Depth Analysis of Crowdfunding Project Descriptions

Abstract

This paper outlines a research methodology based upon Latent Dirichlet Allocation (LDA), an unsupervised machine learning method to extract latent topics from crowdfunding project descriptions and examine the relationship of these topics and project funding success. This paper also analyzed whether sentiment, subjectivity and text positioning of a project content description have an impact on the funding rate. By analyzing 10.000 past crowdfunding projects, this study identified 8 latent topics that can be used in predicting project funding by backers. The findings also indicate that textual content that have a positive sentiment increases the likelihood of funding success. Furthermore, project descriptions that are objective in the middle and at the end have a positive effect on funding success. This study also found that there are interaction effects present with the identified variables that impact the rate of funding depending on their location within a project content description. The analysis presented in this paper can help to improve on the identification of the success factors of entrepreneurial and crowdfunding projects.