Ui exerciser can drive android apps to reach different code branches so that dynamic analysis can be improved. A key problem of it is how to generate valid text inputs, which can satisfy input constraints. Existing works use heuristic rules or machine learning to predefine inputs. However, Many apps restrict inputs based on dynamical feedback from their servers. If the pre-defined inputs don’t satisfy the constraints, these exercisers will be stuck ! Thus, we design a feedback-driven text input exerciser. The key insight is the feedback information shown on UI can guide input generation. We use Machine learning and Natural language processing to parse hints, then use constraint solver to iteratively generate inputs. We combine our tool with existing exercisers and dynamic analysis tools. The results show that our tool can help them to achieve higher code coverage and find several unknown vulnerabilities.