文本主题特征和医疗众筹筹资绩效：基于LDA模型的研究

Abstract

Medical crowdfunding is an important channel to collect money to cover the medical expenses for poor families with severe illness. However, the fundraising performance of medical crowdfunding projects is still at a relatively low level. Therefore, to improve the fundraising performance, it is natural to stimulate more people to offer help. Project descriptions play a key role in disclosing crowdfunding project information and influencing potential donors. Therefore, the impact of crowdfunding project text information on fundraising performance is investigated, based on the data of 110329 medical crowdfunding projects from Qingsong Chou. The Latent Dirichlet allocation model is employed to analyze the potential semantics of project descriptions. Based on Aristotle's rhetorical theory of three persuasive strategies - credit strategy, rational strategy and emotional strategy, six topic variables are defined, that are (1) image-building (credit strategy), (2) disease narratives and (3) medical expenses (rational strategy), (4) family and affection, (5) persuasion and gratitude, and (6) positive narratives (emotional strategy). A linear regression model is built to quantify the impact of different topics on fundraising performance. Meanwhile, model is adopted to estimate the moderating effect of number of topics. The results indicate that topics emphasizing disease, expenses and positive narratives significantly improve the fundraising performance; Topics highlighting family and affection, persuasion and gratitude and image-building have a significantly negative effect on fundraising performance. Furthermore, an increase in the number of topics enhances the negative impact of persuasion and gratitude on fundraising performance, while diminishing the effects of image-building, medical expenses and positive narratives. These findings not only deepen the understanding of the effect of textual features on fundraising, but also provide new ideas for improving fundraising performance.