So far, we have considered only datasets in which there are two kinds of entities: objects and attributes. However, in several settings it is more natural to consider a dataset as having three, or even more, kinds of entities. For example, we have considered word-document matrices derived from emails, and seen how to extract relationships between words and documents. Another possibility, though, would be to build a 3-dimensional matrix indexed by words, documents, and times, whose entries would represent the frequency of words used in documents during given time periods.