Detecting Function Inputs and Outputs for Learning-Problem Generation in Intelligent Tutoring Systems

Designing of the function interface is one of the key skills in programming. That requires feedback, which can be generated in the necessary quantity only by an intelligent tutoring system. In this paper, we propose a method of extracting function descriptions from inline comments in open-source code and find noun phrases that describe the data items passed to and returned from the function. We compare two popular NLP tools for parsing sentences and two different similarity measures to find the best-performing combination and develop sophisticated methods of filtering functions to increase the percentage of correctly marked functions. We achieved correctly marking more than 80% of the automatically selected functions, which significantly speeds up creating banks of learning problems for intelligent tutoring systems in programming learning.