Systems Literature Analysis Engine (2024)

Undergraduates: Hanqi Xiao, Robert Peters, Davyd Voloshyn, Jonathan Baeden Vester

Faculty Advisor: Kristen Lich
Department: Computer Science, Health Policy and Management

---

To predict the ripple effects of interventions in complex systems such as public health, decision-makers need an accurate map of the system as it is and how its parts are related. Currently, these maps are created by humans who engage stakeholders and validate relationships using the scientific literature. However, humans are slow and imperfect at conducting literature reviews, limiting the size and accuracy of system maps.​

We built a modular natural language processing (NLP) pipeline, the Systems Literature Analysis Engine (SLAE), to demonstrate proof of concept that NLP could be used to build a causal loop diagram (CLD) by extracting relationships from scientific literature. SLAE was piloted on a test dataset of papers related to depression, and the baseline model created a structurally sound CLD with low accuracy (5.8%). However, after an NLP contest to crowdsource innovative ideas, model accuracy was improved to 45.8%. This shows the potential of SLAE as a general-purpose technology for synthesizing knowledge into system maps that can be used by decision-makers to create interventions that account for complexity instead of being thwarted by it.