Finding mathematically equivalent or structurally similar problems in a dataset, rather than just keyword-based matching.