{"paper":{"title":"Breaking the Reasoning Horizon in Entity Alignment Foundation Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A parallel encoding strategy lets entity alignment foundation models generalize directly to unseen knowledge graphs.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Kexuan Xin, Wei Hu, Yuanning Cui, Zequn Sun, Zhangjie Fu","submitted_at":"2026-01-29T02:18:45Z","abstract_excerpt":"Entity alignment (EA) is critical for knowledge graph (KG) fusion. Existing EA models lack transferability and are incapable of aligning unseen KGs without retraining. While using graph foundation models (GFMs) offer a solution, we find that directly adapting GFMs to EA remains largely ineffective. This stems from a critical \"reasoning horizon gap\": unlike link prediction in GFMs, EA necessitates capturing long-range dependencies across sparse and heterogeneous KG structuresTo address this challenge, we propose a EA foundation model driven by a parallel encoding strategy. We utilize seed EA pa"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We propose a EA foundation model driven by a parallel encoding strategy... This facilitates anchor-conditioned message passing and significantly shortens the inference trajectory by leveraging local structural proximity instead of global search... Extensive experiments verify the effectiveness of our framework, highlighting its strong generalizability to unseen KGs.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That seed EA pairs are always available as reliable local anchors and that parallel streams plus the merged relation graph can fully capture necessary long-range dependencies without introducing new errors or losing critical global structure in sparse heterogeneous KGs.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A parallel encoding strategy with anchor-conditioned message passing and a merged relation graph allows entity alignment foundation models to generalize to unseen knowledge graphs by shortening the reasoning horizon.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A parallel encoding strategy lets entity alignment foundation models generalize directly to unseen knowledge graphs.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"0ace23ce69b327f4b0c1b9e4f2bf829148ae9c97ac1baf3dedd1c473d875ab0c"},"source":{"id":"2601.21174","kind":"arxiv","version":2},"verdict":{"id":"d744b61b-eadb-4b29-bf92-7352edbf8eba","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T10:07:46.685442Z","strongest_claim":"We propose a EA foundation model driven by a parallel encoding strategy... This facilitates anchor-conditioned message passing and significantly shortens the inference trajectory by leveraging local structural proximity instead of global search... Extensive experiments verify the effectiveness of our framework, highlighting its strong generalizability to unseen KGs.","one_line_summary":"A parallel encoding strategy with anchor-conditioned message passing and a merged relation graph allows entity alignment foundation models to generalize to unseen knowledge graphs by shortening the reasoning horizon.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That seed EA pairs are always available as reliable local anchors and that parallel streams plus the merged relation graph can fully capture necessary long-range dependencies without introducing new errors or losing critical global structure in sparse heterogeneous KGs.","pith_extraction_headline":"A parallel encoding strategy lets entity alignment foundation models generalize directly to unseen knowledge graphs."},"references":{"count":25,"sample":[{"doi":"","year":2025,"title":"Bronstein, ˙Ismail ˙Ilkan Ceylan, Mikhail Galkin, Juan L","work_id":"89e30246-b6d6-4391-abb1-5fdf6f4162bd","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2013,"title":"Translating embeddings for modeling multi-relational data","work_id":"5ea7a693-9cf8-42eb-8b4a-291701366657","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2017,"title":"Neusymea: Neuro-symbolic entity alignment via varia- tional inference","work_id":"92a26edf-263f-41d1-9d9f-15653a65e05a","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"[Chenet al., 2025 ] Zerui Chen, Huiming Fan, Qianyu Wang, Tao He, Ming Liu, Heng Chang, Weijiang Yu, Ze Li, and Bing Qin","work_id":"6a1b912f-bdca-4d80-88bd-2509c295c429","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"A prompt-based knowledge graph foundation model for uni- versal in-context reasoning","work_id":"d36fc62a-71bc-4d21-8366-be4a0cf02964","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":25,"snapshot_sha256":"aaa14357c37c5e045b3a297cde0b8cd90a21ec1016c980b6c057f6a03563af34","internal_anchors":0},"formal_canon":{"evidence_count":1,"snapshot_sha256":"7b73cc5453ae89c29605225a954530802d54f30886d00395464708fac56c71fb"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}