Mining Source Codes of Programming Learners by Self-Organizing Maps

摘要

There are lots of source codes on the Internet, which are resources valuable for programming tutoring. In school programming training, for instance, teachers may want to know how many solutions students could use to solve a given problem and which solution is the most popular. For this reason, we propose a method which is based on self-organizing maps to discover the mainstream solution for each problem by mining its corresponding source codes that learners submitted. We believe that the problem that has many solutions is much suitable for programming learners and that the mainstream solution should be recommended. 1510 source codes submitted by 40 students for 60 problems were mined in our experiment. The results show that each problem has 3.26 solutions on average and that 90% problems have their unique mainstream solutions.