Using machine learning to understand social media discourse on the co-use of tobacco and cannabis

The simultaneous use of tobacco products and cannabis is increasingly common among youth aged 18 to 24 years. Co-use has a deleterious effect on physical and mental health. Social media is being used to promote and discourage co-use. However, there is no comprehensive monitoring or surveillance on how co-use is discussed in these platforms. We conducted this research to analyze the content and sentiments of social media posts about co-use and develop a surveillance system using Deep Learning. We hypothesized that the majority of posts would have a positive sentiment promoting co-use, possibly driven by the rapid increase of tobacco and cannabis use. We entered co-use–related keywords into a social media monitoring platform and analyzed the sentiment data over a specific period. In addition, we developed a codebook with pre-defined features and manually classified the sentiment of a random subset of posts. Results from the manual coding study showed that approximately 3% of the posts had a positive sentiment towards co-use, while 28% had a negative sentiment. Approximately 60% of the posts mentioned physical or mental health effects of co-use. However, despite the harmful health effects, over 90% of the posts did not mention quitting. The manually coded data was used to train and develop a surveillance system with a Long Short-Term Memory (LSTM) classification model. The model achieved an accuracy of 78% on the test data and was further optimized to attain 91% accuracy, making it a potential tool for public health officials and researchers in the future.