TERMCOOCCUR Procedure

The TERMCOOCCUR procedure has two key phases of its calculations. The first phase limits the terms that are considered for analysis. This phase is important for computational reasons. You can control the set of term pairs that are considered in the first phase by using the term selection options listed in Table 2.

When the term pairs meet the criteria of the first phase, the second phase computes the pointwise mutual information (PMI) score for these candidates. For term and term , the PMI score between the terms is given by

where and n is the number of terms in the analysis. When the argument of the log is negative and the log is undefined, the value of 0 is used for the PMI score.

You have control over several tuning adjustments that can be made for this calculation. The SIMILARITY statement enables you to threshold, smooth, and normalize values to get the desired results. For more information, see the explanations of the algorithm tuning options in Table 2.