A community is a group of entities not only with high density but also with similar characteristics. Just recent works have merged the graph structure of the networks with the attributes of nodes. Some of them make weight modification, linear combination, and the most promising are the model-based algorithms, which faces the mixture of structure and attributes in a centralized way, this integration increases the dimensionality of the problem. Moreover, not all the attributes are relevant for that purpose, and there are works that rank attributes in each iteration of the community detection process, others do not consider multiple groups with the same attribute. Based on the principle that linked nodes are likely to adopt similar attributes, we calculated weight to each attribute to get the most RELevant Node Attributes (RELNA) based on the relations of the nodes. We integrated the selected attributes to overlapping community detection methods like the well-known CESNA (Communities from Edge Structure and Node Attributes) and our method QMUCA (Quality Measure to Upgrade Communities using Attributes and structure), which is a model-based approach that uses optimization in a regression analysis and a new similarity measure. Experiments show that QMUCA with the ranking of RELNA improves the performance of the accuracy of other methods, but we are still working on it.