Extracting Creative Procedural Knowledge from Photoshop Tutorials
Federico Madden · Scholar Commons (Santa Clara University) · 2021
For my senior design project, I created a web app that, given the text of an online Photoshop tutorial, uses a neural network to extract specific Photoshop actions (e.g. File - New or Layers - New Layer) that the tutorial either implicitly or explicitly instructs the reader to take. This neural network consists mainly of a BERT embedding layer followed by an LSTM module, with a linear multi-class, multi-label classification layer on top. The network achieved a macro average F1 score of 15.1% on tutorial sentence classification tasks.