I've trained a word2vec model not for English but for an Asian language 'Sinhala'. in the later phase, I'm going to use this trained model to get the sentence similarities in order to detect plagiarism in Sinhala documents. Please explain to me how to measure the accuracy of the trained model.I'm a university student. I have no previous knowledge of these things.
There is no universal measure of word2vec model quality or 'accuracy'.
The commonly-reported "accuracy" is typically based on a set of english-language analogy questions that were used by Google in their original word2vec paper (and included in their source code release). See for example:
https://github.com/tmikolov/word2vec/blob/master/questions-words.txt
To make a similar calculation for another language, you'd need to supply a similar set of evaluation questions for that language. I don't know of any collection of such questions for Sinhalese, or other languages, so you may have to find or create it yourself. (You could create an alternate file in the same format, and use the existing evaluation methods, specifying your alternate file.)