Could you provide some domain independent embedding vaild methods ？

[svjack]

I take a glace at this project, the main idea of this project is to valid the embedding conclusion by labeled data (most are fixed provided by human setting word analogy relations or similar relations) so the embedding and the labeled data "matching" degree is important. When apply it in a specific domain, the simplest methods to use your project is to customize the "labeled" data (words or relations) as my interest. replace it in the project data. So, the future issue of this topic is: are there some methods to replace the "labeled" data and use some meaningful "parts" in train data to construct new evaluate framework ? Or if you can provide some related papers or projects for me to read or hand on, thank you.