On Researchgate, one scholar asked "how to check plagiarism?", and his presentation is as follows:

Presently, there are about 30 answers to this question. Some of them provide the software sites to do plagiarism checking, some of them said they could detect whether the articles plagiarize or not by reading the whole paper. If they found the styles jumping from sentence to sentence or from paragaraphs to paragraphs, then this paper possibly plagiarize some paragraphs from difference sources.

But if one plagiarizes a large part of their contents from one paper and we never read that plagiarized paper before, then how can we detect the plagiarism? In fact, one can only read a limited amount of papers, so it's rarely possible to surely detect a paper's plagiarism just by reading it. Even if you feel the paper has plagiarism, how can you provide the evidence? So software is very important in this aspect.

According to my own working experience, I think software provided by iThenticate is very useful in checking plagiarism.

The following is my answers to the question "how to check plagiarism". 

Our journal (Journal of Mountain Science) will first check the plagiarism of a manuscript by using ITHENTICATE after it is submitted. The plagiarism checking software was incorporated into the online manuscript system. The crosschecking system will compare the similarity of the submitted paper with millions of papers previously published and articles in the internet (usually by google), and provide a separate similarity index with one single paper and the total similarity index with all these papers.
The checked result is displayed in two columns, the left column is the checked paper, the system use different colors to mark the similar contents, and the right column is the list of the possibly plagiarized papers, which are listed in the order of similaritity index percentage.
The marked color in the checked paper is corresponding to the color of the possible plagiarized paper.
We usually set a similarity index for rejecting a paper. When the total similarity index is up to 40%, or the similarity index with one single paper reaches 20%, the manuscript will be directly rejected.
When the similarity index has not reached the set level but is relatively high, we will read the whole paper and check which part is taken from othe papers. If the abstract or conclusion is almost the same with one single paper, we'll definitely reject it too. If most figures and tables are the same with one single paper or the key data are copied from one paper, we'll reject the paper too.
If only the introduction part or discussion part is taken from other papers but not from one or two papers, we'll require the authors to revise the paper, and suggest them organize these two parts in their own way and write them in their own words.

We start to use the iThenticate software from 2010. At first, we found that a relatively high proportion of papers plagiarized some paragraphs from other papers, some even copied a large part of contents from other papers. Now high plagiarism rate can be rarely seen for this year's new submission or there are still plagiarisms in some paragraphs, especially in the Introduction or Discussion parts.