Discovering mainstream knowledge in source codes of programming learners

There are lots of archives of problems for programming practice on the Internet. The problems in these archives, however, are not organized effectively for programming tutoring. To select the problems suitable for students to exercise, it is required to know the knowledge necessary to solve the problems and the knowledge that students have mastered. This can be done by discovering the problem-solving knowledge in the program source codes that the students have submitted. We propose a method based on Formal Concept Analysis (FCA) to discover the knowledge behind source codes of problem solutions. By constructing a formal context from many source codes for a given problem, the FCA is applied to building a formal concept lattice, from which the mainstream knowledge for solving the problem can be discovered. 60 problems were mined by the proposed method, which analyzed 2710 source codes submitted by 51 students. We found that on average each mainstream method had about 50% students employing it while other methods had only about 5%.