Using Unstructured Data to Improve the Continuous Planning of Critical Processes Involving Humans

The success of processes executed in uncertain and changing environments is reliant on the dependable use of relevant information to support continuous planning at runtime. At the core of this planning is a model which, if incorrect, can lead to failures and, in critical processes such as evacuation and disaster relief operations, to harm to humans. Obtaining reliable and timely estimations of model parameters is often difficult, and considerable research effort has been expended to derive methods for updating models at run-time. Typically, these methods use data sources such as system logs, run-time events and sensor readings, which are well structured. However, in many critical processes, the most relevant data are produced by human participants to, and observers of, the process and its environment (e.g., through social media) and is unstructured. For such scenarios we propose COPE, a work-in-progress method for the continuous planning of critical processes involving humans and carried out in uncertain, changing environments. COPE uses a combination of runtime natural-language processing (to update a stochastic model of the target process based on unstructured data) and stochastic model synthesis (to generate Pareto-optimal plans for the process). Preliminary experiments indicate that COPE can support continuous planning effectively for a simulated evacuation operation after a natural disaster.