Unsupervised learning of domain incorporated metaphor vectors
Learn vectors for terms used figuratively across domains. For example, the word wooden used in the domain of books is different from that of kitchen and housewares
Here is a co-relation matrix of the separated words in domain of boooks and kitche, and the terms which are semantically similar to them:
Clearly, they are separated in vector space, and made closer to terms which they are semantically similar to.