JoinGR: Learning to Traverse Join Graphs for Table Retrieval

Sandipan De; Abhijit Chakraborty; Sambaran Bandyopadhyay; Vivek Gupta

cs.CLOct 1, 2026

JoinGR: Learning to Traverse Join Graphs for Table Retrieval

Authors: Sandipan De, Abhijit Chakraborty, Sambaran Bandyopadhyay, Vivek Gupta

Organizations: Arizona State University · Adobe Research, India

Abstract

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.

Figures & tables

Figure 1: Overview of JoinGR table retrieval for the example question. (1) Semantic retrieval selects the top-5 anchor tables based on dense similarity between the query and table/column representations. (2) Starting from these anchors, JoinGR performs query-conditioned multi-hop traversal over the join graph; edge thickness indicates traversal probability and node size reflects aggregated deposit received during propagation. Gold tables are highlighted in orange. (3) Deposits accumulated from all traversal paths are aggregated to produce final table scores.

Figure 2: Example of the join graph constructed from a relational schema. Circular nodes denote key columns, rectangular nodes denote non-key columns, solid edges represent intra-table connectivity, and dashed edges represent inter-table join relationships.

Panel of Figure 3 — Figure 3: Expansion in Join Graph. From frontier table $f$ , JoinGR scores outgoing join edges to neighbor $t_{1}$ , $t_{2}$ and $t_{3}$ , locally normalizes the scores, and propagates deposit.

Dataset	Split	#DB	#Tab.	#Q	#N	#Intra	#Join
Spider	Dev	20	81	1034	441	492	63
Spider	Train	166	876	7000	4503	4704	793
BIRD	Dev	11	75	1534	798	3345	102
BIRD	Train	69	522	9428	3539	4559	418
BEAVER	DW	1	97	121	1530	3536	197

Table 1: Dataset and join-graph statistics. #N denotes column nodes; intra edges connect columns within a table, and join edges connect foreign-key-related columns across tables.

Panel of Figure 4 — Figure 4: Distribution of queries by number of required gold tables.

BGE-large-en-v1.5
Method	P		R		PR
Method	@5	@15	@5	@15	@5	@15
Vanila Retriever	0.352	0.175	0.476	0.697	0.124	0.339
REaR	0.347	0.179	0.474	0.684	0.174	0.364
JoinGR ${}_{\text{H}}$	0.319	0.195	0.423	0.759	0.058	0.405
JoinGR	0.423	0.226	0.553	0.875	0.157	0.645

Table 2: Table retrieval results on BEAVER (DW) . JoinGR

{}_{\text{H}}

is the heuristic version of our algorithm. For JoinGR

{}_{\text{X}\rightarrow\text{Y}}

we utilized the scorer trained on BIRD dataset and evaluated on BEAVER.

BGE-large-en-v1.5
Method	P		R		PR
Method	@5	@15	@5	@15	@5	@15
Vanilla Retriever	0.349	0.127	0.918	0.983	0.830	0.964
REaR	0.359	0.198	0.937	0.985	0.871	0.970
JoinGR ${}_{\text{H}}$	0.300	0.123	0.789	0.951	0.650	0.921
JoinGR	0.360	0.188	0.934	0.973	0.877	0.961

Table 3: Table retrieval results on BIRD . JoinGR

{}_{\text{H}}

is the heuristic version of our algorithm. For JoinGR

{}_{\text{X}\rightarrow\text{Y}}

we utilized the scorer trained on BEAVER dataset and evaluated on BIRD.

Panel of Figure 5 — Figure 5: Precision and Recall @5, 15 by number of gold tables: (a) BIRD, (b) BEAVER (DW).

Panel of Figure 6 — Figure 6: R@10 on BEAVER (DW) for different anchor budgets.

Appendix figures & tables3 assets

Supplementary material from the paper’s appendix.

Appendix

	$\displaystyle\phi(e)=\Big($	$\displaystyle x\,;\,\mathrm{emb}(e)\,;\,\text{MaxSim}(T_{n}),$
		$\displaystyle\text{MaxSim}(T_{f})\,;\,s_{f}\,;\,h\Big)$

Appendix

Algorithm 2 Query-Conditioned Table Search

Hyperparameter	Value
Max BFS hops $H$	$5$
Scorer hidden dim	$64$
Dropout	$0.3$
Optimizer	Adam
Learning rate	$10^{-3}$
Weight decay	$10^{-4}$

Appendix

Table 4: Hyperparameters used in all main experiments.

BGE-large-en-v1.5
Method	P		R		PR
Method	@5	@15	@5	@15	@5	@15
Vanilla Retriever	0.215	0.072	0.745	0.752	0.729	0.741
REaR	0.216	0.099	0.745	0.746	0.733	0.735
JoinGR ${}_{\text{Heuristic}}$	0.209	0.072	0.715	0.747	0.699	0.736
JoinGR	0.213	0.072	0.737	0.747	0.722	0.736

Appendix

Table 5: Table retrieval results on Spider .

Explore similar work

CORE-T: COherent REtrieval of Tables for Text-to-SQL
Jan 19, 2026Hassan Soliman, Vivek Gupta, Dan Roth +1Text-To-Sql Tabular Data
Ubiquitous Knowledge Processing Lab (UKP Lab), Department of Computer Science TU Darmstadt and National Research Center for Applied Cybersecurity ATHENE, Germany · Arizona State University · University of Pennsylvania
TabJoinBench: A Benchmark for Joinable Table Discovery
Sep 30, 2026Sandipan De, Jin Wang, Vivek GuptaData Lakes Benchmark Datasets
Arizona State University
Finding the Right Tables and Columns: A Benchmark and Corpus-Adaptive Embeddings for SQL Schema Retrieval
Jul 14, 2026Qingcheng Zeng, Puxuan Yu, Aman Mehta +2Text-To-Sql Schema
Northwestern University · Snowflake Inc.

Jan 19, 2026, cs.CLJ/K move · Enter open · S save

CORE-T: COherent REtrieval of Tables for Text-to-SQL

Hassan Soliman, Vivek Gupta, Dan Roth, Iryna Gurevych

Ubiquitous Knowledge Processing Lab (UKP Lab), Department of Computer Science TU Darmstadt and National Research Center for Applied Cybersecurity ATHENE, Germany · Arizona State University · University of Pennsylvania

Text-To-Sql Tabular Data Large Language Model Baselines Cache

Open paper AskPDF arXiv

Realistic text-to-SQL workflows often require joining multiple tables. As a result, accurately retrieving the relevant set of tables becomes a key bottleneck for end-to-end performance. We study an open-book setting where queries must be answered over large, heterogeneous table collections pooled from many sources, without clean scoping signals such as database identifiers. Here, dense retrieval (DR) achieves high recall but returns many distractors, while join-aware alternatives often rely on extra assumptions and/or incur high inference overhead. We propose CORE-T, a scalable, training-free framework that enriches tables with LLM-generated purpose metadata and pre-computes a lightweight table-compatibility cache. At inference time, DR returns top-K candidates; a single LLM call selects a coherent, joinable subset, and a two-step additive adjustment stage restores strongly compatible tables. Across Bird, Spider, MMQA, and Beaver, CORE-T improves over DR by up to 22.7 points in table-selection F1 while returning up to 40% fewer tables, and by up to 24.4 points in multi-table execution accuracy, and uses 1.64-4.20x fewer total selection tokens than LLM-intensive baselines.

Abstract

Explore similar work

CORE-T: COherent REtrieval of Tables for Text-to-SQL

TabJoinBench: A Benchmark for Joinable Table Discovery

Finding the Right Tables and Columns: A Benchmark and Corpus-Adaptive Embeddings for SQL Schema Retrieval