{"paper":{"title":"Informative Graph Structure Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"InGSL reduces redundant edges in graph structure learning by balancing similarity and diversity through a mutual-information strategy.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Bingde Hu, Canghong Jin, Can Wang, Da Zhong Li, Hai Lin, Jiawei Chen, Sheng Zhou, Shen Han, Zhiyao Zhou","submitted_at":"2026-05-16T04:46:59Z","abstract_excerpt":"The quality of graph-structured data is fundamental to the success of modern graph analysis techniques such as Graph Neural Networks (GNNs). However, real-world graph data is often suboptimal, suffering from issues such as noise and incomplete connections. Graph Structure Learning (GSL) has emerged as a promising technique that adaptively optimizes node connections. However, we observe that the effectiveness of GSL often comes at the cost of a dramatic expansion in edge count, resulting in significant storage and computational overhead.\n  In this work, we reveal that this limitation stems from"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"InGSL achieves significant performance improvements at a reduced number of edges by jointly considering both similarity and diversity in edge construction by incorporating a mutual-information-guided learning strategy.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that similarity-based edge construction is the primary source of structure redundancy and that a mutual-information term can reliably select informative edges without losing critical connections or adding new computational overhead that offsets the savings.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"InGSL reduces edge redundancy in existing graph structure learning methods by adding a mutual-information-guided diversity term, delivering better results with fewer edges across six tested frameworks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"InGSL reduces redundant edges in graph structure learning by balancing similarity and diversity through a mutual-information strategy.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"e53b6f644f67758173a20d1e9e125995b36ebe067f43584611ae3374dd8079b0"},"source":{"id":"2605.16809","kind":"arxiv","version":1},"verdict":{"id":"928eaf83-998f-4bb7-9c77-b98735d20168","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T21:04:28.461464Z","strongest_claim":"InGSL achieves significant performance improvements at a reduced number of edges by jointly considering both similarity and diversity in edge construction by incorporating a mutual-information-guided learning strategy.","one_line_summary":"InGSL reduces edge redundancy in existing graph structure learning methods by adding a mutual-information-guided diversity term, delivering better results with fewer edges across six tested frameworks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that similarity-based edge construction is the primary source of structure redundancy and that a mutual-information term can reliably select informative edges without losing critical connections or adding new computational overhead that offsets the savings.","pith_extraction_headline":"InGSL reduces redundant edges in graph structure learning by balancing similarity and diversity through a mutual-information strategy."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.16809/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-19T21:31:19.271752Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T21:11:43.493007Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T19:01:56.279734Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T18:33:26.418001Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"17527a542db66185537bb708801e5d053cedaae33ee8d70b49015fb6cd8a2a38"},"references":{"count":56,"sample":[{"doi":"","year":2017,"title":"Kipf and Max Welling","work_id":"b46adfec-f85a-4df1-be97-69900280ef43","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2017,"title":"Neural message passing for quantum chemistry","work_id":"b2b7a819-41b5-4e72-b648-0a1e483424d0","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2016,"title":"Convolutional neural networks on graphs with fast localized spectral filtering.Advances in Neural Information Processing Systems, 29, 2016","work_id":"a9f3ba97-af38-430b-bf54-5c4884fed14b","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Hatllm: Hierarchical attention mask- ing for enhanced collaborative modeling in llm-based recommendation.CoRR, abs/2510.10955, 2025","work_id":"bfc76c09-a5da-44ce-ae9a-91097f6a00ec","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"Simplifying graph convolutional networks","work_id":"33c3dc29-7df5-4a85-8622-0d9d9f2ed6e0","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":56,"snapshot_sha256":"39a18657f5906db1929a781d44d6f9e5f73ee4d73c2b0afd7144f1c724af35a0","internal_anchors":1},"formal_canon":{"evidence_count":2,"snapshot_sha256":"7225c22511d7db6da24c8b6af66982c4f72634d142ce743733490a696794f912"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}