Embodiments provide for predicted data use obligation matching using data differentiators. Example embodiments retrieve a cross-link relationship graph structure from a data repository, where the cross-link relationship graph structure includes a plurality of cross-link relationship graph nodes connected by a plurality of cross-link relationship graph edges. Each cross-link relationship graph node is associated with a unique logical data record set identifier of a plurality of logical data record set identifiers associated with a dataset identifier. For each unique logical data record set identifier, the cross-link relationship graph structure is traversed. Based at least in part on a separation measure associated with each cross-link relationship of one or more cross-link relationships associated with the unique logical data record set identifier, one or more data use obligation scores for the unique logical data record set identifier is generated.