Drugs or Dancing? Using Real-Time Machine Learning to Classify Streamed “Dabbing” Homograph Tweets

摘要

Dabbing is a new and popular method of using marijuana that involves inhaling vapors from heating marijuana concentrates. As the emergence of legal, regulated markets continues in the U. S., it is possible that dabbing marijuana concentrates will gain traction. Dabbing may present new hazards to marijuana users including increased risk of fires from igniting extracts with butane and increased incidence of addiction due to higher concentrations of the psychoactive chemical tetrahydrocannabinol (THC) inhaled when dabbing. Twitter can be used to better understand health behaviors by analyzing conversations around marijuana dabbing, however, collecting relevant tweets is complex given that "dabbing" is also a term used to describe a dance done at sporting events and the process of covering a sneeze. We developed a machine learning algorithm to classify tweets and identify relevant marijuana dabbing (mdab) tweets. We found our classifier to be reliable in differentiating mdab from other dabbing tweets. Machine learning based classifiers have potential for helping public health researchers and practitioners to handle the large volumes of complex Twitter data in order to learn from this new information stream. Our technique, used to solve this particular tweet differentiation problem, is easily applicable to any homograph differentiation problem in tweet space.