Project Description
The objectives of this research are to determine the most effective automated methods for identifying different stages of recovery—such as contemplation, relapse, and active recovery—in social media discussions related to eating disorders (ED). We have created a dataset of thousands of posts and comments downloaded from a social media forum intended to help people with eating disorders. We have developed an annotation guideline for coding the text with labels that indicate the recovery stage they indicate. We will use the guideline to have multiple trained coders label a sample of data and then provide the guideline and sample to a machine learning model optimized to perform automatic coding and ask the model to label the rest of the data. We will then assess the results for accuracy, consistency, and possibly new measures of annotation quality. Overall this work could enable the creation of better automatic tools for searching and summarizing content. This content, or tools that use them, might help ED patients and counsellors to better understand eating disorders and ultimately boost efforts towards recovery from these disorders.
Tasks and Responsibilites
The student tasks include: data labeling, especially in recognizing subtle cues within posts that signal various stages of recovery. The student will also need to participate in training (to learn the annotation process) and in discussions about the results and why they labelled things as they have. The student will participate in the poster session at UWM, and in any publications that are prepared during their time as a SURF fellow.