EAGER: A Holistic Heterogeneous Temporal Graph Transformer Framework with Meta-learning to Combat Opioid Epidemic

The devastating and lethal opioid epidemic has largely been fueled with various opioids in the United States. Unfortunately, driven by considerable profits, opioid trafficking has co-evolved with modern technologies, such as, social media platforms have been utilized for marketing and selling illicit drugs including opioids, which has attracted increasing attention from both public health agencies and law enforcement.

As online opioid trafficking activities are nimble and resilient, it calls for novel techniques to effectively detect opioid trades to facilitate proactive response strategies. By advancing capabilities of machine learning and data science, the goal of this project is to design and develop a holistic framework to model and analyze dynamic multi-modal data to fight against online opioid trafficking and, thus, help combat opioid epidemic.

This research will enable a conceptual framework for the federal and state governments, public health agencies, law enforcement, and local communities to develop proactive strategies to build up a drug-free world - one community at a time.

By engaging novel disciplinary perspectives, this exploratory, yet transformative, high risk-high payoff work will involve radically different approaches for the development of an integrated framework to combat online opioid trafficking. The research will have three key components. First, the team will propose a novel heterogeneous temporal graph (HTG) to comprehensively model and abstract multi-modal posts and relational information over time on social media.

Second, based on the constructed HTG, the research team will develop an innovative graph transformer to learn user representations for opioid trafficker detection. Third, to tackle the challenge of lack of sufficient labeled data for model training, the team will further develop a new meta-learning algorithm by joining unsupervised graph structure and small amount of supervised training data to update the model.

This will enable the model to quickly adapt to a new task, such as identifying a new type of traded opioid and its traffickers on social media, using only a few samples and training iterations. The developed holistic framework for the detection of online opioid trafficking activities will have significant impacts on addressing the critical national opioid epidemic facing our society.

The research will be beneficial to data mining and machine learning communities, as well as multidisciplinary domains such as public health, epidemiology, social and behavioral sciences. The outcomes of this project will be made publicly accessible and broadly distributed. The project will integrate research with education through novel curriculum development, participation of underrepresented groups, and student mentoring activities.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.