This talks about the concept of weighted quasi-identifiers but then it is not like the way we are using, they assign a weight and adjust it to create the equivalence classes

Our notion of weight is a relation of how the clustering has to be performed

Ex, a qid that has a higher weight will determine the ordering of the elements in such a way that tuples with similar domain values of that qid will be grouped together, this determines the precedence of the sorting, or on what parameter the grouping has to be done, our weights are not changed in between the iteration, it is defined and used to calculate the absolute score of the record, after which grouping of the records will be performed and solved

All records in each equivalence class are generalized to be the same with the class center in the class

does this mean all the k records are made the same as the center record and is not generalised as such ?

Does this help solve any problem

Adjustment of feature weights is done in each iteration, but in our method the weights actually represent the correlation between the sensitive attribute and the qid

The higher the correlation the qids and sensitive value have, the greater is the contribution of the qid to determine the sensitive attribute

Shortcomings

In this paper we assume all quasi-identifier features in F are numerical since we emphasize the introduction of this clustering algorithm. Therefore, the evaluation formula of diss(rmn,ain) can be defined as Equation (3). Noted that the details about the dissimilarity evaluation for categorical features can be referred to [10].