SolMiner: mining distinct solutions in programs.

Given a programming problem, because of a variety of data structures and algorithms that can be applied and different tradeoffs, such as space-time, to be considered, there may be many distinct solutions. By comparing his/her solution against others' and learning from the distinct solutions, a learner may quickly improve programming skills and gain experience in making trade-offs. Meanwhile, on the Internet many websites provide venues for programming practice and contests. Popular websites receive hundreds of thousands of submissions daily from novices as well as advanced learners. While these websites can automatically judge the correctness of a submission, none extracts distinct solutions from the submissions and provides them to learners. How to automatically identify distinct solutions from a large number of submissions is a challenging and unresolved problem. Due to diverse coding styles and high programming flexibility, submissions implementing the same solution may appear very different from each other; in addition, dealing with submissions at scale imposes extra challenges. We propose SolMiner, a solution miner, that automatically mines distinct solutions from a large number of submissions. SolMiner leverages static program analysis, data mining, and machine learning to automatically measure the similarity between submissions and identify distinct solutions. We have built a prototype of SolMiner and evaluated it. The evaluation shows that the technique is effective and efficient.