Joining heterogeneous tables 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, while running every candidate method incurs unnecessary cost. We introduce MetaJoin, a label-free framework for adaptive join selection. 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. This allows join selection to adapt after execution rather than committing to a one-shot routing decision. We also introduce MNN Join, a transformation specialist that searches for string transformations by maximizing mutual-nearest-neighbor coverage between two unaligned columns and uses the same reciprocal structure to validate its output. We evaluate MetaJoin on 670 datasets from six benchmarks. It achieves 0.817 macro row-F1, slightly exceeding a benchmark-level oracle that is 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 outperforms every evaluated transformation baseline. Finally, 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.
Figures & tables
Figure 1. The MetaJoin workflow: (1) input table pair, (2) join-column selection, (3) evidence-guided routing over the specialist portfolio, and (4) validation and repair before committing the final join.
Operator
Example
Operator
Example
Substring / character extraction (char)
auto-split, keep field
a:b → b
drop leading char(s) †
Xenon → enon
drop trailing char(s) †
Smiths → Smith
drop first char
_id → id
drop last char
id_ → id
remove 2nd char
u.s → us
remove 2nd-to-last char
abXc → abc
take leading prefix
Robert → Rob
strip parenthetical
Ann (NE) → Ann
strip leading numeric
12. Ann → Ann
Table 1. The MNN Join operator library Ω : 25 unary string operators, listed two per row. † repeatable within one search step.
Figure 2. (a) Matching behaviors present in each benchmark, manually characterized from gold pairs. (b) Median and maximum source–target pair space.
Method
Auto-Join
FlashFill
SyGuS
PROSE
AutoFJ
Magellan
Macro
Tokens ( a+b )
External baselines
DeepJoin
.81 ± .000
.83 ± .000
.80 ± .000
.69 ± .000
.17 ± .000
.00 ± .000
.55 ± .000
0
Lotus
.69 ± .001
.83 ± .000
.79 ± .001
.62 ± .001
.49 ± .001
.50 ± .003
.65 ± .000
7k
FM-Wrangle
.34 ± .000
.21 ± .000
.11 ± .000
.16 ± .000
.52 ± .000
.50 ± .000
.31 ± .000
31k
GXJoin
.23 ± .000
.22 ± .000
.34 ± .000
.25 ± .000
.11 ± .000
.00 ± .000
.19 ± .000
0
Auto-Join-PBE
.36 ± .000
.07 ± .000
.04 ± .000
.12 ± .000
.12 ± .000
.00 ± .000
.12 ± .000
0
Table 2. Label-free row- F1 (mean ± std, three seeds; shaded by value), the benchmark macro mean, and mean LLM tokens per task ( a : controller, b : specialists; 0 : no model calls). Bold : best per column.
Figure 3. Per-dataset capability of each matching class. For each class, we bin the row- F1 of its strongest standalone specialist as easy ( ≥.80 ), challenging ( .50 – .80 ), hard ( .20 – .50 ), or not able ( <.20 ).
Figure 4. Standalone components and signals. (a) Input-only certificate column selection versus individual signals on Auto-Join . (b) Reciprocal covmnn coverage versus raw idfcos similarity. (c) Sensitivity to the underlying similarity metric.
Figure 5. Controller mechanism ablations over three seeds. Blue reports the four transformation-dominant benchmarks ( 609 datasets), with affected share and median gain among affected datasets. Gold reports AutoFJ ( 50 tables), with help/hurt counts at ∣Δ∣>0.05 . Green reports Magellan ( 11 tables), with affected-table counts at the same threshold.
Figure 6. Portfolio sensitivity. MetaJoin is compared with the strongest standalone method retained after removing specialists from the portfolio.
Figure 7. Cost and scalability. (a) End-to-end wall-clock time on the four transformation-dominant benchmarks. (b) Transformation-search time versus target size on a log–log scale. (c) MNN Join ’s raw reciprocal search on synthetic tables.
Figure 8. Full-system backbone swap. Each fully- M configuration replaces every swappable LLM component with backbone M , including the LLM-based transformation specialist.
Join discovery aims to identify tables from large data repositories that can augment a query table with complementary information, enabling downstream tasks such as data exploration, feature engineering, and business intelligence. Although numerous join discovery methods have been proposed, existing studies rely on method-specific benchmark construction, making reproducible and fair comparison difficult. We present TabJoinBench, a benchmark for evaluating join discovery methods across semantic, relational, and hybrid data lake scenarios. TabJoinBench constructs query-candidate pairs using source-specific validation strategies, systematically introduces structural, representation, and semantic changes through composable perturbations while preserving reliable ground truth. We evaluate representative join discovery methods spanning set-based, feature-based, and learned approaches, together with general-purpose language-model embedding baselines, and publicly release the processed datasets, ground-truth annotations, and generation pipeline to facilitate reproducible evaluation and future research.
Retrieving the right tables is a prerequisite for Text-to-SQL over realistic databases. Dense table retrievers rank schema elements independently, but this ignores a key source of evidence: some required tables are not mentioned in the question and become identifiable only through their join relationships to already relevant tables. We introduce JOINGR, a join-aware table retrieval method that treats the database join graph as the retrieval space. Columns are represented as graph nodes, while intra-table and foreign-key relationships are represented as typed edges. Given a question, JOINGR selects semantically similar anchor tables, traverses join edges with a query-conditioned scorer, and aggregates the resulting edge deposits into table scores. The scorer is a lightweight MLP on top of frozen query, node, and edge embeddings, trained with a pairwise margin loss over gold tables. On BIRD and Spider datasets, JOINGR is competitive with the strongest retrieval baselines. On BEAVER, a challenging enterprise benchmark with multi-hop table requirements, JOINGR substantially improves recall over dense retrieval and re-ranking baselines. Cross-domain experiments show that the learned scorer transfers across benchmarks, indicating that the method captures reusable joingraph traversal behavior.
Sandipan De, Abhijit Chakraborty, Sambaran Bandyopadhyay +1
Integrating unstructured data into relational database systems is increasingly important as demand grows for natural language querying and analysis. A semantic join, joining two tables under a natural-language predicate, can be evaluated with a large language model (LLM), but comparing every pair of tuples requires O(M x N) LLM invocations and is cost-prohibitive at scale. Existing systems reduce this cost but typically commit to a single fixed strategy (e.g., embedding similarity or one batched scheme) regardless of the data or the join predicate. We propose an LLM-agent-based decision pipeline that optimizes semantic joins by matching the execution strategy to the characteristics of the underlying tables. An LLM advisor routes each join to one of two strategies: a Cluster Join, which prunes candidates via unsupervised embedding clustering and sample-based filtering, or a Classifier strategy for predicates that reduce to a shared discrete label set. Across three diverse datasets (IMDb reviews, email contradictions, and Stack Overflow tags), the advisor consistently identifies the optimal execution strategy for each workload. This dynamic routing proves decisive: it outperforms adaptive block join (ABJ) by 20-33 F1 points across all datasets while consuming fewer tokens on two of the three, and achieves higher F1 scores than featurized-decomposition join (FDJ) at one to two orders of magnitude lower token cost.