Publicación:

Hydra: Funding State Prediction for Kickstarter Technology Projects Using a Multimodal Deep Learning

Resumen

Since crowdfunding started, thousands of entrepreneurs have presented their projects to the public to fund them. During the 2009–2019 period, 37% of all Kickstarter projects, one of the most popular crowdfunding platforms, were successfully funded. Different Machine Learning algorithms have been used, considering all the categories in this platform to develop predictive models. However, their research works only reached 20% for the Technology category. The main aim of this study is to develop a Multimodal Deep Learning model with three layers: a Multilayer Perceptron for metadata, a Convolutional Neural Network for project descriptions, and a Bidirectional LSTM model for backers comments. This proposal can predict funding state of Technology projects on Kickstarter. In order to train the model, we created a dataset with 27K Technology projects on Kickstarter between 2009 and 2019. The performance of this proposal reached an AUC value of 93%. Thus, the problem was solved with a different approach that combinates different types of networks to improve results.