Term-Term correlations suggest how often two terms appear in the same chunks or, to put it another way, in the same semantic networks.
The closer the cosine similarity approaches 1.0, the more closely related the two terms are in terms of co-occurrence across all the chunks of the corpus.
Term-Term correlations can be used to construct concept newtwork graphs.