1. Though automated item gernerationg sound promising and attractive, it is actually very chanllenging. As stated in the first paragraph, "if the cognitive processes involved in solving the test items are known", which is neally impossible. However, it is crutial in automated item generation, since it is related directly to define the radicals and incidentals features that influcence the difficulties of the items. In other words, the beta in Equation 4 are never satisfactory.
2. How to specify design matrix in Equation 4 need much investiagation, juas as the specification of Q matrix in CDM models.
3. It was noted that if all varaince of item parameters are 0 (solid line in Figure 1), which indiate all item parameters are equal to their family mean and family information equals item information. If the covariance was nonzero but treated as unrelated mistakenly, the amount of information will be overestimated. How can we find the results in Figure 1? It was not investigated in simulations.