{"paper":{"title":"Decoupled and Divergence-Conditioned Prompt for Multi-domain Dynamic Graph Foundation Models","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"DyGFM decouples semantic and temporal patterns in dynamic graphs and uses divergence-conditioned prompts to enable effective multi-domain pre-training without negative transfer.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Haonan Yuan, Jianxin Li, Junhua Shi, Philip S. Yu, Qingyun Sun, Xingcheng Fu","submitted_at":"2026-05-13T13:50:03Z","abstract_excerpt":"Dynamic graphs are ubiquitous in real-world systems, and building generalizable dynamic Graph Foundation Models has become a frontier in graph learning. However, dynamic graphs from different domains pose fundamental challenges to unified modeling, as their semantic and temporal patterns are inherently inconsistent, making the multi-domain pre-training difficult. Consequently, the widely used \"pretrain-then-finetune\" paradigm often suffers from severe negative knowledge transfer. To the best of our knowledge, there exists no multi-domain dynamic GFM. In this work, we propose DyGFM, a Dynamic G"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"DyGFM consistently outperforms 12 state-of-the-art baselines on both node classification and link prediction tasks, achieving superior effectiveness and efficiency.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that semantic-temporal decoupling plus divergence-aware expert selection will reliably prevent negative transfer across arbitrary domains without introducing new biases or requiring extensive hyperparameter tuning per domain pair.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"DyGFM introduces decoupled pre-training and divergence-conditioned prompts to create the first multi-domain dynamic graph foundation model that outperforms baselines on node classification and link prediction.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"DyGFM decouples semantic and temporal patterns in dynamic graphs and uses divergence-conditioned prompts to enable effective multi-domain pre-training without negative transfer.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"0ec6571c7fa281cf22c3946cf8230904259a2c4e2e4e3c031c0547afa51b4bfe"},"source":{"id":"2605.13540","kind":"arxiv","version":1},"verdict":{"id":"8315ce0c-0a13-4545-b3d4-f17c1e715768","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T20:18:30.424750Z","strongest_claim":"DyGFM consistently outperforms 12 state-of-the-art baselines on both node classification and link prediction tasks, achieving superior effectiveness and efficiency.","one_line_summary":"DyGFM introduces decoupled pre-training and divergence-conditioned prompts to create the first multi-domain dynamic graph foundation model that outperforms baselines on node classification and link prediction.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that semantic-temporal decoupling plus divergence-aware expert selection will reliably prevent negative transfer across arbitrary domains without introducing new biases or requiring extensive hyperparameter tuning per domain pair.","pith_extraction_headline":"DyGFM decouples semantic and temporal patterns in dynamic graphs and uses divergence-conditioned prompts to enable effective multi-domain pre-training without negative transfer."},"references":{"count":122,"sample":[{"doi":"","year":2002,"title":"Random graph models of social networks,","work_id":"d1a7ac2e-1bbe-4116-9417-a74108fce148","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"Graph neural networks for social recommendation,","work_id":"5b2cba34-653a-4e9f-81c0-b2003455ae89","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Graph neural networks for friend ranking in large-scale social platforms,","work_id":"86b6b116-f9cb-41ed-924b-99a9af6187d9","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Graph convolutional neural networks for web-scale recommender systems,","work_id":"4fa665f5-4881-4183-b3f3-c7ef4d9340eb","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"Session-based recommendation with graph neural networks,","work_id":"ab77aff2-4abf-4b7d-ad28-d3bd7a463e9c","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":122,"snapshot_sha256":"67f01850ce2c6f723006ab0b4d2f5e5f4f324c8bcf0451147af0490639e9a2d7","internal_anchors":5},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}