Author: Joan Francesc Serracant Lorenzo
Supervisor: Coloma Ballester
Presentation time: 8:30 h
Virtual Room: 2.1 | Live presentation URL
Nonverbal language plays a role when entrepreneurs pitch their business ideas to potential investors. Some authors have proposed that several crucial clues from nonverbal communication can affect on accessing early stage investments and even long-term firm survival. The aim of this project is to apply recent supervised deep learning strategies to this field of investigation.Our proposal leverages from a recent work  where the authors gathered a new dataset from entrepreneurial pitching sessions recorded on video and found a set of nonverbal language characteristics that strongly correlate to investment outcome and start-up performance in early stages. We use their dataset to extract skeletal information of speakers and learn patterns in human body poses that correlate with those nonverbal language characteristics. Inspired by the state of the art techniques in human action recognition, we propose a method that uses Spatial-temporal Graph Convolutional Networks, adapted to perform a regression instead of classification, given the nature of our data. The obtained results exhibit good performance with a mean average error of1.7%. Due to the lack of reference results on our custom dataset, we implemented other traditional machine learning techniques with hand-crafted features to compare our results with.