MetaJoin, a label-free framework for adaptive join selection, was introduced in a preprint posted to arXiv on October 12. The framework adapts its approach after execution rather than committing to a one-shot routing decision.
The method matters because joining heterogeneous tables often requires different forms of correspondence, including exact agreement, syntactic transformation, semantic similarity, and entity-level matching. The appropriate join method is often unknown in advance, and running every candidate method can incur unnecessary costs.
MetaJoin uses input evidence to choose an initial specialist, then uses specialist-specific output evidence to determine whether that result should be retained, compared with an alternative, escalated, or repaired. The preprint describes the method and its evaluation. The framework was evaluated on 670 datasets from six benchmarks, achieving a macro row-F1 score of 0.817. This slightly exceeds a benchmark-level oracle that was given the best standalone specialist for each benchmark (0.812) and outperforms label-free selection policies built from the same specialist portfolio by more than 0.13. Across the 609 datasets in the four transformation-dominant benchmarks, MNN Join, a transformation specialist introduced alongside MetaJoin, outperforms every evaluated transformation baseline. The preprint also notes that MetaJoin's accuracy varies by only 0.015 as the controller backbone ranges from 3B to 14B parameters, showing that output validation and repair substantially reduce dependence on the initial routing decision.
The preprint does not include independent test results or peer review.