If the documents where themselves stored in a database they have and id and the contents. The clustering key (an index) would be on the id. It's inverted because the contents are deconstructed into tokens with a list of ids that contain that token. Now the contents (tokens) server as the indexed value.