RENSA: Rich Environment Metadata to Navigate Shared and Distributed Endpoints for Automated Federated SPARQL Query Generation

Victor Eiti Yamamoto; Takeda Hideaki; Yamamoto Yasunori

Source-linked AI summary

RENSA: Rich Environment Metadata to Navigate Shared and Distributed Endpoints for Automated Federated SPARQL Query Generation

Victor Eiti Yamamoto, Takeda Hideaki, Yamamoto Yasunori

arXiv:2608.28963v1Aug 29, 2026cs.DBcs.CL

TL;DR

Federated SPARQL query generation lacks the detailed structural metadata needed for efficient source selection across decentralized datasets, while existing approaches may rely on runtime ASK queries. RENSA extends SBM with class and URI authority profiles to infer variable constraints and assign sources statically. On LargeRDFBench, it achieved comparable source-selection outcomes with zero additional requests, while class and authority inference reached average F1-scores of 0.672 and 0.831.

Problem
Decentralized datasets often lack fine-grained structural metadata, and existing metadata standards miss interlinks and authority distributions needed for federated source selection.
Method
RENSA extends SBM with class relationships and URI authority distributions to infer semantic constraints and assign related query patterns to endpoints using static profiles.
Results
RENSA achieves zero additional requests across all queries, comparable source-selection quality, and average F1-scores of 0.672 for classes and 0.831 for authorities.
Takeaways & Limitations
RENSA supports source selection and federated plan generation without runtime ASK verification, while grouping contiguous query paths into cohesive endpoint execution blocks.
Takeaways & Limitations
The evaluation did not explicitly measure total query execution time, and generated mappings are logically correct but not necessarily execution-cost optimal.

Abstract

from arXiv · show

The number of knowledge graph databases has increased significantly with the proliferation of knowledge graph technologies. Knowledge graphs enable the dynamic integration of distributed data through federated SPARQL queries. However, constructing efficient queries in a federated environment is challenging due to the lack of detailed structural knowledge across decentralized datasets. While standards like VoID provide basic metadata, they often fail to capture the complex interlinks and authority distributions necessary for optimization. Consequently, current engines frequently rely on runtime ASK queries for source selection, increasing communication overhead. We propose RENSA, a federated SPARQL query generation framework that leverages an extension of SPARQL Builder Metadata (SBM). By integrating class and authority information, mapping subject and object usage to specific predicates, RENSA enables precise source selection and semantic constraint inference for query variables without runtime communication. The generated profiles represent less than 1\% of the original dataset triples in most cases, ensuring storage efficiency. Evaluation on the LargeRDFBench benchmark (13 datasets with >1B triples, 32 queries) shows that RENSA achieves source selection results comparable to state-of-the-art methods while eliminating ASK query overhead. Furthermore, we demonstrate that RENSA infers class and authority constraints for query variables, enabling the identification of data sources even across heterogeneous endpoints. These profiles additionally offer human-readable structural insights for semi-automated query generation.

1 Introduction

Federated SPARQL query generation is hindered by opaque decentralized data structures and metadata that misses interlinks and authority distributions. RENSA addresses these gaps with extended profiles, static source selection, and semantic constraint inference, achieving comparable selection quality without runtime ASK overhead.

Motivation: Efficient federated queries require structural knowledge of classes, predicates, and interlinks that decentralized environments often leave opaque.Publishers may omit structural metadata because maintaining it requires substantial manual effort and expertise.
Motivation: VoID provides basic statistics but misses complex interlinks and authority distributions, leading engines to use runtime ASK queries for source selection.These runtime operations introduce communication overhead and can degrade performance as query volume grows.
Approach: RENSA extends SPARQL Builder Metadata with URI authority analysis and cross-endpoint link detection for federated query optimization.Its integrated stages profile class relationships and authority distributions, infer variable constraints, and assign endpoint mappings through branch-and-prune planning.
Approach: Static profiles group related patterns into cohesive SERVICE blocks, eliminating runtime ASK queries during source selection.This design prioritizes optimization before execution and reduces remote interactions.
Evaluation: On LargeRDFBench, spanning 13 datasets and over one billion triples, RENSA achieved comparable source-selection precision while bypassing runtime ASK overhead.The evaluation also reports accurate class and authority constraint prediction for query variables.

2 Related Work

Federated source-selection methods use either runtime verification or precomputed metadata, but existing metadata approaches leave structural and provenance gaps. RENSA combines class-to-class structure with URI authority information to support static selection and variable-constraint inference.

Existing approaches: Runtime-based source selection uses ASK queries to test source relevance but introduces substantial communication overhead in federated environments.Metadata-based methods reduce this dependence by matching triple-pattern classes and properties against precomputed descriptions.
Existing approaches: Existing metadata-based engines may still use ASK queries for unbounded predicates or patterns fixing only a subject or object.Other systems index URI authority information but do not jointly provide the structural and provenance representation used by RENSA.
RENSA: RENSA combines structural class-to-class profiling with URI authority analysis to perform source selection statically.Its hybrid profile supports semantic constraint inference and optimized federated plans without runtime verification.
RENSA: RENSA jointly encodes class-level structure, URI authority distributions, sample triples, and crawl metadata for static selection and query-variable constraint inference.The comparison is summarized in Table 1.

3 Methodology

RENSA uses an extended SBM-based profile to capture dataset structure, authority distributions, and class relations, then applies semantic inference and endpoint assignment to generate federated plans.

RENSA Profiling: The RENSA Profile combines schema and statistical information with class-class relationships and URI authority distributions for decentralized SPARQL endpoints.It extends SBM and related metadata standards to describe links across endpoint boundaries.
RENSA Profiling: ClassRelation metadata records subject classes, predicates, object classes or datatypes, sample triples, and operational crawl information.These elements support query verification and profile traceability.
RENSA Profiling: LDCs have explicit local type declarations, whereas LUCs appear in local triples without explicit types and may be externally defined or schema-unknown.The distinction identifies entities whose schema is defined elsewhere.
RENSA Profiling: The profile summarizes six LDC/LUC relationship categories, including internal schema links, local attributes, and potential incoming or outgoing inter-dataset links.These summaries help infer data presence when explicit schema information is sparse.
RENSA Profiling: URI authority relations record authority URIs, matching triple counts, and subject or object roles, enabling identification of entities from other endpoints.Authorities are extracted as common prefixes to discover potential cross-endpoint links.
Source Selection and Query Generation: RENSA parses queries into abstract syntax trees that preserve operator nesting, extracts triple patterns, and uses profile lookups to infer feasible classes and authorities for variables.The targeted implementation covers a SPARQL 1.1 fragment including BGPs, JOIN, UNION, and OPTIONAL.

4 Evaluation

The evaluation measures RENSA’s source-selection efficiency and semantic-inference accuracy on a simulated federation using LargeRDFBench query sets and benchmark-derived ground truth.

Experimental Setup: RENSA was evaluated on 13 Virtuoso 7.2 endpoints hosting LargeRDFBench datasets in a Docker Compose federation.Experiments used an Apple M1 Max with 64GB RAM, Python 3.11, and rdflib.
Experimental Setup: Three query sets were tested: Simple S1–S14, Complex C1–C10, and Large-scale L1–L8.The query groups cover progressively different benchmark settings.
Ground Truth: Ground truth for each variable’s expected class and authority was constructed from the corresponding values in LargeRDFBench query results.All-literal variables were assigned rdfs:Literal and no authority; URI variables received extracted authorities.
Evaluation Metrics: Source-selection efficiency is measured by selected sources (#T) and additional ASK requests (#AR), while semantic inference uses Precision, Recall, and F-measure.Inference metrics are evaluated against the constructed ground truth.

5 Results and Discussion

RENSA produces compact metadata for static source selection and evaluates its effectiveness across source selection and semantic inference. It achieves zero additional ASK requests, competitive source counts, and generally high inference accuracy, while retaining limitations in execution-time measurement and generic-query over-selection.

Metadata Compression: Less than 0.3% of the original data volume is represented by RENSA metadata in most cases, with SWDFood an exception at 13.2%.The profiles are larger than VoID descriptions because they encode class relationships and authority distributions.
Source Selection Effectiveness: #AR = 0 across all queries because RENSA performs source selection entirely from pre-generated metadata.This avoids runtime ASK requests during source selection.
Source Selection Effectiveness: 103 sources are selected for complex queries, compared with 104 for CostFed, 106 for HiBISCuS, and 111 for ANAPSID.For large-scale queries, RENSA selects 86 sources, compared with 98 for CostFed and 118 for HiBISCuS.
Class and Authority Inference: Authority inference achieves an overall average F1-score of 0.831, while average recall is 0.735 for classes and 0.893 for authorities.Many queries reach perfect scores of 1.0; lower scores occur with generic patterns, schema heterogeneity, or entities having multiple types.
Class and Authority Inference: RENSA can remove an inferred class when it produces an empty authority set, relying on URI authority to avoid prematurely pruning sources.This fallback broadens selection when class information may be incomplete.
Limitations: The evaluation did not explicitly measure total query execution time, and generated queries are not necessarily execution-cost optimal.The framework also sometimes over-selects sources for generic queries or hub datasets such as DBpedia.

6 Conclusion

RENSA combines structural and namespace-level metadata to support precise source selection and semantic constraint resolution without runtime ASK queries. Its results are encouraging, while future work targets temporal benchmarking, over-selection, and stronger optimization.

Conclusion: RENSA enables precise source selection without runtime ASK queries by integrating URI authority information with RDF class metadata.The approach uses RENSA Profiling to provide structural awareness for federated optimization.
Conclusion: 0.672 and 0.831 are the average F1-scores for semantic inference of classes and authorities, respectively.These scores indicate that structural and namespace-level metadata provide signals for resolving variable constraints.
Conclusion: RENSA profiles support identification of contiguous paths and cohesive join blocks through local optimization at remote endpoints.The profiles also assist data exploration and more effective query formulation.
Future work: Future work will benchmark temporal aspects, investigate hierarchical authority profiling and Bloom-filter instance summaries, and incorporate cost-based path pruning.These directions aim to quantify performance gains, mitigate over-selection, and improve optimization or integration with existing methods.