This project uses machine learning emotion classification processes in order to create emotionally representative data from Project Gutenberg public domain novels. The data is then fitted into a visualization that is further rendered into book covers for the analyzed pieces. Both proprietary and hand-crafted APIs are used to analyze emotions found in the corpuses of the 25 high-ranking books from Project Gutenberg. Multiple emotion recognition models are evaluated for accuracy and value within the literary space. The machine learning models are then compared to human evaluations of emotions, based on existing psychoanalytic research. Results indicate that machine learning models are able to adequately predict the emotions in novels and generate supporting book cover art.
The final outcome of this project will be printable manuscripts with custom-generated covers. The visualizations are built using D3js and data storytelling techniques. The charts will be imported into InDesign for final production and typesetting. In order to avoid copyright issues, all books used are in the public domain, with original publish dates between 1800 and 1925. The physical book will contain the text of the original author, as well as accompanying text and diagrams provided from this project.