Our work addresses the generation of first-order constraints to reduce symmetries and improve the solving performance for classes of instances of a given combinatorial problem. To this end, we devise a model-oriented approach obtaining positive and negative examples for an Inductive Logic Programming task by analyzing instance-specific symmetries for a training set of instances. The learned first-order constraints are interpretable and can be used to augment a general problem encoding in Answer Set Programming. This extented abstract introduces the context of our work, contributions and results.