Abstract
In this paper, we propose a novel method to measure similarity between program source codes. Different to others, our method doues not compare two source codes directly but compares two reference vectors, where a reference vector is calculated from one source code and a set of reference source codes. This means that our method requires no original source code when considering an application open to public such as a search engine for the program source code on the internet. We have built a simple search system and have evaluated with Java source codes made in the university course of basic programming. Results show that the system can achieve quite high average precision rate in very short time which means the proposed method can measure correct similarity very fast.
