To support researchers during this scientific workflow, we introduce an anytime algorithm for the bottom-up discovery of generalized multimodal graph patterns in knowledge graphs. Each pattern is an interconnected set of triples with special variables for classes, datatypes, and value patterns. Upon discovery, the patterns are automatically converted to SPARQL queries and presented in an interactive facet browser together with metadata, graph visualizations, and provenance information, enabling researchers to explore and analyse the patterns as well as reproduce and share relevant data selections. We evaluate our method both quantitatively and qualitatively, on social historical data from The Netherlands, and with the help of domain experts in social science and the humanities.