Due to smart word processors and powerful Web-searching engines, lots of plagiarism prevail, especially in digital texts. So it is very crucial to develop efficient and effective anti-plagiarism tools to prevent or identify document plagiarism. Till now, a few plagiarism detecting systems have been announced. All previous plagiarism detection studies focus on how to measure the similarity of documents. In this paper, we propose a new approach to reconstruct the evolution process of suspected texts in order to detect plagiarized documents. For this, we propose two major metrics: spatial plagiarism similarity and temporal plagiarism similarity. And by combining these two similarity measure, we give conclusively the evolutionary plagiarism probability model by adopting the Weibull distribution, which is one of extreme distribution used to compute the statistical significance of genomic sequence matching. The main difference of our model to the previous studies is that our model can estimate the plagiarism and its direction as a temporal event. An experiment with a group Internet-posted news clearly coincided to the real plagiarism among those news.