Explainability and graph learning from social interactions