Structure is the Signal: Graph Encodings and GNNs for Constraint Repair in Collaborative KGs. Vázquez, M., Innerebner, K., Prock, A., Klambauer, G., Lex, E., Schimunek, J., & Polleres, A. In 23rd European Semantic Web Conference (ESWC), May, 2026. to appear![Structure is the Signal: Graph Encodings and GNNs for Constraint Repair in Collaborative KGs [pdf]](https://bibbase.org/img/filetypes/pdf.svg "pdf")
Paper abstract bibtex Collaborative Knowledge Graphs (KGs) rely on evolving, ``soft'' constraints, making automated repair a challenge for rigid symbolic methods. In this paper, we propose a structure-aware neural framework that learns to repair violations directly from historical edit patterns. We perform a systematic study of graph encoding strategies, specifically comparing flattened predicate-as-node representations against multi-relational graphs equipped with a custom dual-representation module. We benchmark these encodings across different GNN backbones and feature initializations. Our experiments show that the Multi-Relational GIN yields the most robust performance, surpassing strong symbolic baselines by over 29% in historical fidelity and achieving a 93% validity rate. Furthermore, ablation studies indicate that topological features, such as role embeddings and edge attributes, are significant performance drivers, often outweighing semantic text features. These findings suggest that effective repair depends heavily on precise topological modeling, reinforcing the premise that structure is the signal.
@inproceedings{vazq-etal-2026ESWC,
title={Structure is the Signal: Graph Encodings and {GNN}s for Constraint Repair in Collaborative {KG}s},
note={to appear},
booktitle={23rd European Semantic Web Conference (ESWC)},
month=may,
day={10--14},
url={http://polleres.net/publications/vasq-etal-2026ESWC.pdf},
year=2026,
author={Miguel V{\'a}zquez and Kevin Innerebner and Alexander Prock and G{\"u}nter Klambauer and Elisabeth Lex and Johannes Schimunek and Axel Polleres},
abstract = {Collaborative Knowledge Graphs (KGs) rely on evolving, ``soft'' constraints, making automated repair a challenge for rigid symbolic methods. In this paper, we propose a structure-aware neural framework that learns to repair violations directly from historical edit patterns. We perform a systematic study of graph encoding strategies, specifically comparing flattened predicate-as-node representations against multi-relational graphs equipped with a custom dual-representation module. We benchmark these encodings across different GNN backbones and feature initializations. Our experiments show that the Multi-Relational GIN yields the most robust performance, surpassing strong symbolic baselines by over 29\% in historical fidelity and achieving a 93\% validity rate. Furthermore, ablation studies indicate that topological features, such as role embeddings and edge attributes, are significant performance drivers, often outweighing semantic text features. These findings suggest that effective repair depends heavily on precise topological modeling, reinforcing the premise that structure is the signal.},
}
Downloads: 0
{"_id":"cyG4G53zxSPN6rRF3","bibbaseid":"vzquez-innerebner-prock-klambauer-lex-schimunek-polleres-structureisthesignalgraphencodingsandgnnsforconstraintrepairincollaborativekgs-2026","author_short":["Vázquez, M.","Innerebner, K.","Prock, A.","Klambauer, G.","Lex, E.","Schimunek, J.","Polleres, A."],"bibdata":{"bibtype":"inproceedings","type":"inproceedings","title":"Structure is the Signal: Graph Encodings and GNNs for Constraint Repair in Collaborative KGs","note":"to appear","booktitle":"23rd European Semantic Web Conference (ESWC)","month":"May","day":"10–14","url":"http://polleres.net/publications/vasq-etal-2026ESWC.pdf","year":"2026","author":[{"firstnames":["Miguel"],"propositions":[],"lastnames":["Vázquez"],"suffixes":[]},{"firstnames":["Kevin"],"propositions":[],"lastnames":["Innerebner"],"suffixes":[]},{"firstnames":["Alexander"],"propositions":[],"lastnames":["Prock"],"suffixes":[]},{"firstnames":["Günter"],"propositions":[],"lastnames":["Klambauer"],"suffixes":[]},{"firstnames":["Elisabeth"],"propositions":[],"lastnames":["Lex"],"suffixes":[]},{"firstnames":["Johannes"],"propositions":[],"lastnames":["Schimunek"],"suffixes":[]},{"firstnames":["Axel"],"propositions":[],"lastnames":["Polleres"],"suffixes":[]}],"abstract":"Collaborative Knowledge Graphs (KGs) rely on evolving, ``soft'' constraints, making automated repair a challenge for rigid symbolic methods. In this paper, we propose a structure-aware neural framework that learns to repair violations directly from historical edit patterns. We perform a systematic study of graph encoding strategies, specifically comparing flattened predicate-as-node representations against multi-relational graphs equipped with a custom dual-representation module. We benchmark these encodings across different GNN backbones and feature initializations. Our experiments show that the Multi-Relational GIN yields the most robust performance, surpassing strong symbolic baselines by over 29% in historical fidelity and achieving a 93% validity rate. Furthermore, ablation studies indicate that topological features, such as role embeddings and edge attributes, are significant performance drivers, often outweighing semantic text features. These findings suggest that effective repair depends heavily on precise topological modeling, reinforcing the premise that structure is the signal.","bibtex":"@inproceedings{vazq-etal-2026ESWC,\ntitle={Structure is the Signal: Graph Encodings and {GNN}s for Constraint Repair in Collaborative {KG}s},\nnote={to appear},\nbooktitle={23rd European Semantic Web Conference (ESWC)},\nmonth=may,\nday={10--14},\nurl={http://polleres.net/publications/vasq-etal-2026ESWC.pdf},\nyear=2026,\nauthor={Miguel V{\\'a}zquez and Kevin Innerebner and Alexander Prock and G{\\\"u}nter Klambauer and Elisabeth Lex and Johannes Schimunek and Axel Polleres},\nabstract = {Collaborative Knowledge Graphs (KGs) rely on evolving, ``soft'' constraints, making automated repair a challenge for rigid symbolic methods. In this paper, we propose a structure-aware neural framework that learns to repair violations directly from historical edit patterns. We perform a systematic study of graph encoding strategies, specifically comparing flattened predicate-as-node representations against multi-relational graphs equipped with a custom dual-representation module. We benchmark these encodings across different GNN backbones and feature initializations. Our experiments show that the Multi-Relational GIN yields the most robust performance, surpassing strong symbolic baselines by over 29\\% in historical fidelity and achieving a 93\\% validity rate. Furthermore, ablation studies indicate that topological features, such as role embeddings and edge attributes, are significant performance drivers, often outweighing semantic text features. These findings suggest that effective repair depends heavily on precise topological modeling, reinforcing the premise that structure is the signal.},\n}\n\n","author_short":["Vázquez, M.","Innerebner, K.","Prock, A.","Klambauer, G.","Lex, E.","Schimunek, J.","Polleres, A."],"key":"vazq-etal-2026ESWC","id":"vazq-etal-2026ESWC","bibbaseid":"vzquez-innerebner-prock-klambauer-lex-schimunek-polleres-structureisthesignalgraphencodingsandgnnsforconstraintrepairincollaborativekgs-2026","role":"author","urls":{"Paper":"http://polleres.net/publications/vasq-etal-2026ESWC.pdf"},"metadata":{"authorlinks":{}},"downloads":0},"bibtype":"inproceedings","biburl":"www.polleres.net/mypublications.bib","dataSources":["gixxkiKt6rtWGoKSh"],"keywords":[],"search_terms":["structure","signal","graph","encodings","gnns","constraint","repair","collaborative","kgs","vázquez","innerebner","prock","klambauer","lex","schimunek","polleres"],"title":"Structure is the Signal: Graph Encodings and GNNs for Constraint Repair in Collaborative KGs","year":2026}