Many people struggle to interpret their dreams and the meaning of various symbols or elements. Creating a tool that automatically codes specific elements of dreams can present an opportunity to highlight significant themes that may provide insight into an individual dreamer’s unconscious. This project trains a model to determine emotions present in a text dream account using a dataset containing dream reports that was used by researchers who wrote a paper titled Our dreams, our selves: automatic analysis of dream reports. The researchers sourced the dream data, a collection of 20k+ dream reports, from dreambank.net. The dataset contains journal-like text accounts of dreams from a number of individuals that have been coded using the Hall/Van de Castle dream coding system. This system was developed by psychologists as a method for doing quantitative content analysis on dreams. It assigns quantitative values to several dream elements: characters (male/female, animal, family, etc), aggression or friendliness of interactions, negative/positive emotions. The dataset also includes information on the dreamers’ profiles and dates of dreams.
For this machine learning project, I used a Random Forest classifier to train and test a model to classify text-based dream data. My process for doing this is as follows: