{"paper":{"title":"Beyond Individual Mimicry: Constructing Human-Like Social network with Graph-Augmented LLM Agents","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"GraphMind augments LLMs with graph learning so social bots can build human-like global network structures and evade detection.","cross_cats":["cs.AI"],"primary_cat":"cs.SI","authors_text":"Chuxuan Zhang, Haoran Bu, Hui Pang, Litian Zhang, Xi Zhang, Zhanyuan Liu","submitted_at":"2026-03-31T09:10:55Z","abstract_excerpt":"Driven by large language models (LLMs), social bot can autonomously engage in local interactions, whose human-like behaviors enable them to evade social bot detection. However, while these botnets exhibit realistic local social interactions, they fail to preserve human-like social network. This is because LLM-based bots are graph-unaware and cannot coordinate over global interactions, which makes those botnets vulnerable to graph neural network (GNN)-based detection. To address this limitation, we propose GraphMind, which equips LLM-driven social bots to explicitly learn and fit human-like soc"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experiments on datasets derived from GraphMind-Botnet show that both text-based and graph-based detection models show substantially degraded performance in distinguishing.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That graph augmentation of LLMs can reliably produce social networks statistically indistinguishable from real human ones at global scale, with no details on fitting metrics, validation datasets, or controls for overfitting to specific network properties.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"GraphMind equips LLM agents with graph awareness to construct human-like social networks, producing botnets that substantially degrade performance of both text-based and graph-based detectors.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"GraphMind augments LLMs with graph learning so social bots can build human-like global network structures and evade detection.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"7a50eb4b480997efeee2d0bd6b157721ce98c13810ab4eb3f2de02ea3c51fe5c"},"source":{"id":"2605.12512","kind":"arxiv","version":1},"verdict":{"id":"18f6034a-8f60-405a-bfce-312e5a894122","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T21:15:50.365549Z","strongest_claim":"Experiments on datasets derived from GraphMind-Botnet show that both text-based and graph-based detection models show substantially degraded performance in distinguishing.","one_line_summary":"GraphMind equips LLM agents with graph awareness to construct human-like social networks, producing botnets that substantially degrade performance of both text-based and graph-based detectors.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That graph augmentation of LLMs can reliably produce social networks statistically indistinguishable from real human ones at global scale, with no details on fitting metrics, validation datasets, or controls for overfitting to specific network properties.","pith_extraction_headline":"GraphMind augments LLMs with graph learning so social bots can build human-like global network structures and evade detection."},"references":{"count":86,"sample":[{"doi":"","year":2025,"title":"Mgtab: A multi-relational graph-based twitter account detection benchmark , author=. Neurocomputing , pages=. 2025 , publisher=","work_id":"00cf0f25-50fe-435c-aca8-4a6c37b21058","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Kronecker graphs: an approach to modeling networks. , author=. Journal of Machine Learning Research , volume=","work_id":"c1d3b2ea-8787-401d-a1c7-5be6f62566fa","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Generating large scale-free networks with the Chung--Lu random graph model , author=. Networks , volume=. 2021 , publisher=","work_id":"12b18678-61e3-42ec-9847-edc22cc4e84f","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Lora: Low-rank adaptation of large language models. , author=","work_id":"77a54ad5-3d2c-4a27-a32e-f22d02d1c18f","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"OpenAI blog , volume=","work_id":"31dc92c3-2fc3-432b-8d63-b7ee13f53a9c","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":86,"snapshot_sha256":"4910909c56229e2737c83952e4c379d3a67e7d31471e1f561f497c20c16af710","internal_anchors":8},"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"}