{"record": {"aliases": ["Profile-Graph Memory for LLM Agents: Implicit Cross-Entity Traversal through Narrative Profiles"], "artifacts": [{"id": "arxiv:2607.19359", "kind": "paper", "url": "https://arxiv.org/abs/2607.19359"}, {"id": "gh:shengtongzhu/prograph", "kind": "repo", "url": "https://github.com/ShengtongZhu/ProGraph"}], "authors": ["Shengtong Zhu"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2607.19359", "source_url": "https://arxiv.org/abs/2607.19359"}, "categories": ["General AI", "Language & Knowledge", "cs.AI"], "description": {"en": "A multi-hop long-term memory benchmark for LLM agents: 1,000 questions at hop depths 1–5 across 10 social-network scenarios, with per-hop evidence annotations."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2607.19359", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2607.19359", "title": "MemHop"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2607.19359", "modality": null, "name": "MemHop", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"admission_policy_version": "2026-09-09.1", "confidence": "Medium", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-09-25T04:40:22Z", "data_status": "primary-source-indexed", "display_eligible": "true", "origin_record_id": "bm_profile-graph-memory-for-llm-agents-implic_7b34345b", "origin_source": "arxiv", "recognition_confidence": "0.7", "record_sha256": "5f505266cc2708803455ca86afac3df85151f94a23031b5053e1effefd37d664", "relation": "introduces", "review_model": "claude-code-default", "review_state": "ai-reviewed", "reviewed_at": "2026-09-24T16:50:59.454540Z", "source_url": "https://arxiv.org/abs/2607.19359"}, "publisher": null, "released": "2026-06-01", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2607.19359", "source_url": "https://arxiv.org/abs/2607.19359"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2607-19359", "source": "claire_radar", "source_benchmark_id": "2607.19359", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": ["Profile-Graph Memory for LLM Agents: Implicit Cross-Entity Traversal through Narrative Profiles"], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Language & Knowledge", "attentionForecast": {"confidence": "Low", "horizon": "7d", "reason": "Agent memory is an active area, but the benchmark comes from a method paper by a small team.", "score": 35}, "capabilities": [], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-24T16:50:59.454540Z", "inputHash": "6406de5d68e9489032908a9fb639fb4053220a72b7a81696b2bf26a312488730", "model": "claude-code-default"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "we introduce MemHop, a multi-hop memory benchmark of 1,000 questions", "canonicalNameSource": "abstract", "decisionReason": "Named benchmark with a fixed question set and hop structure. It reports percentage scores that compare systems, and the data is publicly released with baselines.", "evaluationEvidence": {"reuse": "Code and MemHop benchmark: https://github.com/ShengtongZhu/ProGraph", "scoring": "ProGraph averages 80.1% on MemHop (matching the FullContext reference)", "task": "a multi-hop memory benchmark of 1,000 questions at hop depths 1-5 across 10 social-network scenarios, with per-hop evidence annotations"}, "model": "claude-code-default", "reviewedAt": "2026-09-24T16:50:59.454540Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "A multi-hop long-term memory benchmark for LLM agents: 1,000 questions at hop depths 1–5 across 10 social-network scenarios, with per-hop evidence annotations.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["benchmark term in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "First, we introduce MemHop, a multi-hop memory benchmark of 1,000 questions at hop depths 1-5 across 10 social-network scenarios, with per-hop evidence annotations."}, "familyId": "bmf_a6d8164cee5a", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_profile-graph-memory-for-llm-agents-implic_7b34345b", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/ShengtongZhu/ProGraph", "data": null, "paper": "https://arxiv.org/abs/2607.19359", "pdf": "https://arxiv.org/pdf/2607.19359", "project": null, "report": "https://arxiv.org/abs/2607.19359"}, "metrics": [], "motivation": "Long-term memory is essential for LLM agents that interact across sessions, yet current memory benchmarks primarily evaluate single-hop recall, leaving multi-hop association largely unmeasured.", "name": "MemHop", "oneLine": "A multi-hop long-term memory benchmark for LLM agents: 1,000 questions at hop depths 1–5 across 10 social-network scenarios, with per-hop evidence annotations.", "paperTitle": "Profile-Graph Memory for LLM Agents: Implicit Cross-Entity Traversal through Narrative Profiles", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2607.19359", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [], "publishers": [{"name": "ShengtongZhu (GitHub)", "organizationType": "community", "role": "benchmark-publisher", "sourceUrl": "https://github.com/ShengtongZhu/ProGraph"}], "readiness": "Runnable", "recognitionConfidence": 0.7, "relation": "introduces", "releasedAt": "2026-06-01", "reviewLifecycle": {"state": "pending"}, "source": {"authors": ["Shengtong Zhu"], "categories": ["cs.AI"], "id": "2607.19359", "title": "Profile-Graph Memory for LLM Agents: Implicit Cross-Entity Traversal through Narrative Profiles", "type": "arxiv", "url": "https://arxiv.org/abs/2607.19359"}, "sourceUpdatedAt": "2026-06-01T09:48:16Z", "topics": ["cs.AI"], "venueAttempts": [], "whyItMatters": "Most agent-memory benchmarks test single-hop recall. MemHop targets multi-hop association and reports comparable accuracy for several memory systems."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
