A framework for distributed representations of domain embedding

In many information retrieval and natural language processing tasks, it is common to use pertained word embedding to alleviate training difficulty. This motivate us to learn low-dimension representations for all domains on the Internet. There are hundreds of millions of domains on the Internet, majority of the domains attribute to only one or two specific field. Representing these property of domains in a low-dimensional and interpretable space attribute to many tasks, such as domain recommendation, domain classification, web search and et al. In this paper, we propose a novel algorithm named domain embedding, an unsupervised model which learns a fixed length representation for each domain. Experimental results show the superior performance of the proposed method.