Tracing subtle patterns: early detection of Anorexia in social media using contextualized BERT embeddings and deep learning
Abstract
In a world where the usage of social media is prevalent, it is important to acknowledge
the fact that these social platforms have been increasingly used to share private
matters with the public eye, which may sometimes reveal underlying symptoms of
diseases unknown to the user. Mental health disorder is a severe issue and with the
growing popularity of these social media platforms, online bullying has also been on
the rise, leading to users being stressed mentally and developing various mental disorders.
In this paper, we tried to evaluate four state-of-the-art deep learning models:
LSTM, BiLSTM, GRU, and BiGRU, along with BERT, to detect eating disorderrelated
content, specifically anorexia, in online communities. Using a large-scale
dataset extracted from Reddit, our research mainly focused on the early detection
of anorexia from the posts/comments of users. We aimed to use the aforementioned
deep learning models in such a way that would surpass the performance of other
models that were used on our dataset and also, result in a significant advancement
in the field of NLP. Our approach involved an extensive analysis of linguistic cues,
semantic context, and user-generated content to identify both subtle and explicit
mentions of anorexia in social media texts. We tried to attain a higher positive
recall and detect anorexia faster than the models used on the dataset by incorporating
the latest techniques and combining various elements from existing approaches
with the best performance in detecting indicators of this potentially life-threatening
eating disorder. With our models, we were able to achieve a high positive class
recall of 0.93, and our best values of ERDE5 and ERDE50 were 12.63% and 6.99%
respectively.