The last several years show that the theft of programming source code grew up to become a serious problem. Possibilities of recognition if a source code is authentic or it was stolen, mainly using plagiarism, already exists but this computing methods searching for similarities are still not correct enough [2]. The aim of this contribution is to use a cluster analysis as a tool for data mining and use it for establishing at least a partial algorithmic approach for plagiarism searching [3, 5]. The most used types of plagiarism search algorithms works in a way they compare a lot of text sections between each other [1, 4]. However, even with several optimizations they are vulnerable to source code customization where syntax is changed but semantics stays the same. Our approach wants to nd a semantic meaning in those text sections and compare said meanings instead of text sections alone. Our proposed method could significantly help evaluating student assignments and to create an informatics courses teaching methodology at the university.