Different methods are proposed to generate the schema of data mart. We propose a new one that starts by clustering the OLAP requirements that are represented as schemas, then, it generates the schemas of the data mart one for each cluster, and finally it validates them. In this work, we will focus on the first step and we will propose a new algorithm “aK-Mode” to cluster our schemas. Indeed, many techniques exist, they are applied to numerical and categorical data, but they cannot be used in our case since we have to take into consideration the semantic aspect while making the comparison. The new algorithm extends the k-mode by modifying the dissimilarity measure.