Precisely understanding and capturing runtime requirements is the basis for the success of self-adaptive systems. Runtime goal models are an effective way to represent such runtime requirements on top of traditional design-time goal models. However, due to the increasingly complex environments and unpredictable stakeholder behaviors, the runtime requirements are challenging to explore at the design time fully and are usually captured manually. In this paper, we present ongoing work on automatically mining runtime requirements from runtime user interactions. Specifically, our approach takes design-time goal models as input, which are first associated with user operation sequences. Then, we mine runtime behavior patterns from the operation sequences, shedding light on the runtime requirements. We have demonstrated our approach with a real system, which is an online data-flow diagram (DFD) modeling platform and has been used for more than three years. By applying our approach, a runtime goal model for this system has been automatically established, reasonably showing modelers’ runtime requirements when modeling DFDs.