Abstract:
Conversations on social media websites like reddit can provide a wealth of supplemental data and information on certain phenomena. However, the unstructured nature of text data means using a variety of natural language processing (NLP) tools to convert information into a more structured format. In this presentation, we walk through the process of training a Named Entity Recognition (NER) model to extract substances and effects from over 3 million comments on two opioid related subreddits. The resulting structured data is then used to identify which substances and effects cooccur in the text and help us understand how people self-medicate while going through opioid withdrawal.