BPMN task instance streaming for efficient micro-task crowdsourcing processes

The Business Process Model and Notation (BPMN) is a standard for modeling and executing business processes with human or machine tasks. The semantics of tasks is usually discrete: a task has exactly one start event and one end event; for multi-instance tasks, all instances must complete before an end event is emitted. We propose a new task type and streaming connector for crowdsourcing able to run hundreds or thousands of micro-task instances in parallel. The two constructs provide for task streaming semantics that is new to BPMN, enable the modeling and efficient enactment of complex crowdsourcing scenarios, and are applicable also beyond the special case of crowdsourcing. We implement the necessary design and runtime support on top of CrowdFlower, demonstrate the viability of the approach via a case study, and report on a set of runtime performance experiments.