{"paper":{"title":"PersonalAI 2.0: Enhancing knowledge graph traversal/retrieval with planning mechanism for Personalized LLM Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"PAI-2 improves LLM factual accuracy through adaptive graph traversal and planning on knowledge graphs.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Alexander Kharitonov, Alina Bogdanova, Artyom Sosedka, Ekaterina Lisitsyna, Evgeny Burnaev, Ilia Perepechkin, Matvey Iskornev, Mikhail Belkin, Mikhail Menschikov, Ruslan Kostoev, Victoria Dochkina","submitted_at":"2026-05-13T13:06:30Z","abstract_excerpt":"We introduce PersonalAI 2.0 (PAI-2), a novel framework, designed to enhance large language model (LLM) based systems through integration of external knowledge graphs (KG). The proposed approach addresses key limitations of existing Graph Retrieval-Augmented Generation (GraphRAG) methods by incorporating a dynamic, multistage query processing pipeline. The central point of PAI-2 design is its ability to perform adaptive, iterative information search, guided by extracted entities, matched graph vertices and generated clue-queries. Conducted evaluation over six benchmarks (Natural Questions, Triv"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"PAI-2 achieves 4% average gain by LLM-as-a-Judge across four benchmarks, reflecting its effectiveness in reducing hallucination rates and increasing precision. Use of graph traversal algorithms gain superior results compared to standard flatten retriever on average 6%, while enabled search plan enhancement mechanism gain 18% boost compared to disabled one by LLM-as-a-Judge across six datasets.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the reported benchmark gains from the planning and traversal mechanisms will generalize beyond the six evaluated datasets and the specific LLMs tested, and that LLM-as-a-Judge reliably measures factual correctness without introducing its own biases.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"PAI-2 improves factual correctness in LLM answers by 4% on average across benchmarks using adaptive graph traversal and planning, with 6% gains from traversal algorithms and 18% from enabled planning.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"PAI-2 improves LLM factual accuracy through adaptive graph traversal and planning on knowledge graphs.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"129c7ff8251de1d7a3888b27ca64190c532dff92044e0c5e68d5481dccdee9d1"},"source":{"id":"2605.13481","kind":"arxiv","version":1},"verdict":{"id":"131a56eb-59a9-433c-8d2e-e97284a29d59","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T19:11:58.043032Z","strongest_claim":"PAI-2 achieves 4% average gain by LLM-as-a-Judge across four benchmarks, reflecting its effectiveness in reducing hallucination rates and increasing precision. Use of graph traversal algorithms gain superior results compared to standard flatten retriever on average 6%, while enabled search plan enhancement mechanism gain 18% boost compared to disabled one by LLM-as-a-Judge across six datasets.","one_line_summary":"PAI-2 improves factual correctness in LLM answers by 4% on average across benchmarks using adaptive graph traversal and planning, with 6% gains from traversal algorithms and 18% from enabled planning.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the reported benchmark gains from the planning and traversal mechanisms will generalize beyond the six evaluated datasets and the specific LLMs tested, and that LLM-as-a-Judge reliably measures factual correctness without introducing its own biases.","pith_extraction_headline":"PAI-2 improves LLM factual accuracy through adaptive graph traversal and planning on knowledge graphs."},"references":{"count":40,"sample":[{"doi":"","year":2025,"title":"Qwen3 technical report, 2025","work_id":"25520164-3a3c-4e1d-859b-33ac0664fb0a","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"DeepSeek-AI, Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, Damai Dai, Daya Guo, Dejian Y ang, Deli Chen, Dongjie Ji, Erha","work_id":"6c91e37a-9316-4c7b-b73c-d24458f40dde","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Glm-4.5: Agentic, reasoning, and coding (arc) foundation models, 2025","work_id":"92c05c04-1be1-4ce0-8e20-fa4d484d5753","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2026,"title":"Wikontic: Constructing Wikidata-aligned, ontology-aware knowledge graphs with large language models","work_id":"62698cbe-c7e2-4038-9524-3de1c0ce9b5b","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Autoschemakg: Autonomous knowledge graph construction through dynamic schema induction from web-scale corpora, 2025","work_id":"e85beb3a-1aaf-4a0b-a1e4-7c1e31f975f6","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":40,"snapshot_sha256":"7f2da35eaa59b8126b7745601aa458094cac082ff213adf30f29c1398d4c296f","internal_anchors":3},"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"}