Crowdfunding Success Prediction using Project Title Image and Convolutional Neural Network

Prediction of crowdfunding success is a challenging problem that has great importance for project creators and platforms. Although meta features, e.g., number of updates or backers, are widely used for success prediction, they are limited to time period after project posting where project creators cannot adapt their profiles. Because of that, ability to predict campaign success in pre-posting phase would significantly improve chance for project success. According to the theory, mostly used methods in this situation are those based on text features, while methods based on the influence of image modality on project success are rare. Due to this, in this article we propose deep learning-based method for crowdfunding success prediction in pre-posting phase using project title image. Experimental results show that image modality could be used for campaign success prediction. Proposed method obtains results comparable to competing methods from literature, but using only one image per campaign and no derived features. It is also shown that deeper convolutional neural network achieves better prediction performance.