TRIGGER WARNING: This article will discuss personal experiences with eating disorders.
Introduction
More than 9% of the global population will have an eating
disorder at some point over their lifetime. About 26% of this population will attempt suicide, and only 10% will receive treatment. So where do the
other 90% go?
When we are affected by untreated diseases, it is common for us to look for support wherever we can. For many of those
struggling with eating disorders, this supportive space is on the internet, in eating disorder communities. The concept
of medical support communities is not new; in fact, a few are remarkably high-functioning, such as online communities
for chronic pain and cancer. In some eating disorder communities, however, there exists a subculture of pro-ana
(pro-anorexia) thinking. Users in these spaces comfortably trade harmful dieting tips and medical misinformation.
Although most social media platforms have attempted to moderate these spaces, pro-ana websites continue to operate under
the radar as message boards, blogs and Q&A threads where users can post about their experiences anonymously.
Surprisingly, existing research is divided on whether pro-ana communities are harmful or helpful and, in turn, whether
people concerned for those affected by EDs should be intervening or not. Some studies of pro-ana groups have labeled
them as empowering spaces for pro-ana
“practitioners” who gain a positive sense of selfhood through their pro-ED advocacy. These researchers argue that
pro-ana groups are transgressive, as young members resist cultural critiques of their bodies. Some studies also
highlight the importance of pro-ana communities for providing social
support. Researchers have used these findings to suggest that clinicians intervene in pro-ana sites by actively
participating in them. They propose that by accurately answering common medical questions, clinicians could prevent
patients from worsening their disorders. Others, however, have described pro-ana sites as spaces which proliferate an
incredibly toxic subculture. They cite pro-ana
communities as an “urgent issue,” and they provide guidelines for clinicians, young people, parents, and educators on
how to address it. This rigid labeling, however, does not fully capture the complexity of interactions that occur
between members of these communities.
Eating disorders can affect anyone, and these communities include
the perspectives of people with different genders, ages and cultural backgrounds. So how can we as researchers
encompass a group with such diverse experiences? In short, we need to go beyond simplistic labels and try to
understand what it is that members of these communities really need. Rather than condemning these groups as
harmful or celebrating them as welcoming, we should be asking: how can our understanding of interactions within
these groups inform the areas of focus for health care providers, peers, family members and those who are struggling
with an eating disorder?
While there has been some ethnographic research and manual content analysis aimed at understanding the experiences of
pro-ana community members, results are limited in scale, and it requires a longer time to conduct a comprehensive study.
For my research, I have used text mining approaches to carry out a large-scale comparative study on a dataset of
50,591 posts.
Corpus and Methods
I gathered data from three distinct pro-ana communities: MyProAna, Eating Disorder Central, and EDAnonymous on Reddit.
These are widely seen by outsiders and users as pro-ana due to the substantial amount of pro-ana content that circulates
on them. Because I wanted to compare communities of different forms, I selected these three by referring to previous
studies, online threads, and outbound links from inactive communities.
👉
It is worth noting that there is some disagreement around the definition of “pro-ana” sites. In fact, one study even cited MyProAna as a recovery community. However, for the purpose of this research and based on my own readings of the text, I will be referring to these three communities as pro-ana. Using Selenium and the Reddit API, I created a web scraper to retrieve the text from each post, as well as associated
metadata such as the username, number of likes, and post time. I collected documents spanning from July 2018 to November
2022 in order to represent recent topics of discussion.
After cleaning the data, I implemented a topic modeling algorithm,
which helps to identify the main topics within a corpus of texts. Topic modeling is an unsupervised machine learning
method that, by performing iterations of probability assignments, determines how likely it is that
a) a given occurence of a word belongs to each topic and
b) a given topic is associated with each document.
In order to implement the model, I had to determine the optimal number of topics. To find this number, I tested the
model on 5, 10, 15, 20, and 30 topics, and I determined that 10 topics produced the most distinct and representative
groups. In the end, the output of the model is a list of the most probable words for each topic, sorted in descending
order, as well as the proportional weight of each topic associated with each document. Because the model is
unsupervised, however, it cannot find meaning in these lists of words that define a topic. For the most comprehensive
results, we need to combine computational methods with our own manual analysis. So, for each topic, I read its 20 most
representative posts and assigned a label based on my interpretation of the texts.
Here’s a walk-through of the process.
