Networks, representing attitudinal survey data, expose the structure of opinion-based groups. We make use of these network projections to identify the groups reliably through community detection algorithms and to examine social-identity-based groups. Our goal is to present a method for revealing polarization and opinion-based groups in attitudinal surveys. This method can be broken down into the following steps: data preparation, construction of similarity-based networks, algorithmic identification of opinion-based groups, and identification of important items for community structure. We assess the method's performance and possible scope for applying it to empirical data and to a broad range of synthetic data sets. The empirical data application points out possible conclusions (i.e. social-identity polarization), whereas the synthetic data sets mark out the method's boundaries. Next to an application example on political attitude survey, our results suggest that the method works for various surveys but is also moderated by the efficacy of the community detection algorithms. Concerning the identification of opinion-based groups, we provide a solid method to rank the item's influence on group formation and as a group identifier. We discuss how this network approach for identifying polarization can classify non-overlapping opinion-based groups even in the absence of extreme opinions.