Abstract: During the process of crowd funding, the initiators commonly face a dilemma to choose between integrity and confidentiality of project's idea disclosure. They need to promote the project as specific as possible for one side and leave some in reserve to avoid ideas giving away for the public. Generally, the initiators release the project progress to the investors at the right time by phases through the text updates on the platform. Therefore, the impacts of periodic text updates are investigated on the successful rate of crowd funding campaigns. Text hierarchical clustering is used to classify the updated text of the crowdfunding project. Euclidean distance function d(x,y)=$\sqrt {\sum\limits_{i = 1}^n {{{({x_i} - {y_i})}^2}} } $ is employed to calculate the distance between words that extracted from the update content. Then according to the distance d_max(C_i, C_j)=$\mathop {\max }\limits_{\mathop \to \limits_{{x_i} \in {C_{i,}}} \mathop \to \limits_{{x_j} \in {C_j}_,} } $ distance($\overrightarrow {{x_i}} ,\overrightarrow {{x_j}} $) to maximize the distance between clusters.The aim is to minimize the distance in the cluster and maximize the distance between clusters. Firstly, the update text and extract keywords are pre-processed for topic analyzing. Six categories of topics are obtained through clustering topics of the updated text using hierarchical clustering method, that are (1) progress reports, (2) content updates, (3) reward related content, (4) time reminders, (5) thanks-related content and (6) social promotion respectively. An econometric model Success_i=α+U'_i.β+Z'_i.γ+ε_i is built to estimate the periodic impact of linguistic cues to updates on successful crowdfunding campaigns. Using 126 593 crowdfunding projects from Kickstarter as research data, with the total number of 407,582 updates, of which 243,730 published during the financing period, accounting for about 60%, while 163,852 updates are generated after fund raising duration. Then, for the three different phases (the early phase, the middle phase and the late phase), the effect of each update topic on the funding results is explored. Since there are some differences between investors' concern between projects, it is examined which update topic should be focused at different phases for each category of the projects. Meanwhile, the different impacts of internal updates and external updates of the update topics on the funding success are tested. Overall, frequent updates improve the successful funding ratio significantly. More specially, it does not exist significant effects in the first phase of funding, while it is much more effective in the middle phase and the late phase. For the update topics, all update topics are positively related to the funding success, but time reminders, thanks related content and progress reports are more powerfully than other three topics. For different categories, the entrepreneurs should focus on updating different topics. The update strategies on experience-related projects and life-related projects are basically the same, while the update strategies of art-related projects are totally different. This study enriched the research of Internet financial and text mining, and provided new perspectives for researches and practices in crowdfunding. It is demonstrated that phased signal transmission plays a role in online financing, which makes up for the deficiency of existing theoretical research. In addition to E-commerce and other trading areas, the investors in online financing will also consider the phased signal transmission, and we set a fire for future theoretical research and practice.

Key words: crowd funding, information updates, financing effect, text mining, topic classification, investment intention