PMI-IR

The PMI-IR (Point wise Mutual Information - Information Retrieval) algorithm employs the technology of a search engine, such as Google Page Rank, or Yahoo, (Krikos, et al. 2005; Kraft et al. 2006)) to extract the frequency of searched keywords within a collection of documents. In general, the algorithm takes as input a word and a set of alternative terms for that specific word. The output is the selection of the terms whose meaning is the closest to the given word. That is to say, the algorithm finds the synonyms by analyzing the co-occurrences of the terms with the given keyword and among them.

This is exactly the case for tagging systems, where we have a collection of contents labeled with different words representing keywords for that collection and we would like to group words having the same meaning.