An Approach for Detecting Student Perceptions of the Programming Experience from Interaction Log Data

Student perceptions of programming can impact their experiences in introductory computer science (CS) courses. For example, some students negatively assess their own ability in response to moments that are natural parts of expert practice, such as using online resources or getting syntax errors. Systems that automatically detect these moments from interaction log data could help us study these moments and intervene when the occur. However, while researchers have analyzed programming log data, few systems detect pre-defined moments, particularly those based on student perceptions. We contribute a new approach and system for detecting programming moments that students perceive as important from interaction log data. We conducted retrospective interviews with 41 CS students in which they identified moments that can prompt negative self-assessments. Then we created a qualitative codebook of the behavioral patterns indicative of each moment, and used this knowledge to build an expert system. We evaluated our system with log data collected from an additional 33 CS students. Our results are promising, with F1 scores ranging from 66% to 98%. We believe that this approach can be applied in many domains to understand and detect student perceptions of learning experiences.