{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:BPPUG7ZVE4WJBRJIH4L5HZVESZ","short_pith_number":"pith:BPPUG7ZV","canonical_record":{"source":{"id":"2605.14401","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-14T05:38:24Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"e86ef8f2ae665fde5c754939c76dc6ce142ff8832a7d5fd7690fa085862edf3e","abstract_canon_sha256":"156daef7a2fed65ad764a16f6963bfa8b54637036ce4bdf6d0d17da8de80a9a3"},"schema_version":"1.0"},"canonical_sha256":"0bdf437f35272c90c5283f17d3e6a49651a6f971ad8c64623f01634207375faf","source":{"kind":"arxiv","id":"2605.14401","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.14401","created_at":"2026-05-17T23:39:07Z"},{"alias_kind":"arxiv_version","alias_value":"2605.14401v1","created_at":"2026-05-17T23:39:07Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14401","created_at":"2026-05-17T23:39:07Z"},{"alias_kind":"pith_short_12","alias_value":"BPPUG7ZVE4WJ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"BPPUG7ZVE4WJBRJI","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"BPPUG7ZV","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:BPPUG7ZVE4WJBRJIH4L5HZVESZ","target":"record","payload":{"canonical_record":{"source":{"id":"2605.14401","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-14T05:38:24Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"e86ef8f2ae665fde5c754939c76dc6ce142ff8832a7d5fd7690fa085862edf3e","abstract_canon_sha256":"156daef7a2fed65ad764a16f6963bfa8b54637036ce4bdf6d0d17da8de80a9a3"},"schema_version":"1.0"},"canonical_sha256":"0bdf437f35272c90c5283f17d3e6a49651a6f971ad8c64623f01634207375faf","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:39:07.482474Z","signature_b64":"OIBFxrFQWvlBOXaWb1YFh4zpoEpr0jA/Dd0n/f946mybH5V39EUdrfDGsK8tlVWdZ8n7miF6np7ehzVUl1KDCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0bdf437f35272c90c5283f17d3e6a49651a6f971ad8c64623f01634207375faf","last_reissued_at":"2026-05-17T23:39:07.481733Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:39:07.481733Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.14401","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:07Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"y3NvNL7kal/n0AN87dXjq9Mfhj059QdV2Nk9w0xEwcaA3jfo8Cd0o5lSLypjqyN2CrAKATt/LoQSpXk+yZ89Dw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-28T23:47:42.347938Z"},"content_sha256":"ff6558c82b766e2b5cad3f2135eba9e7aea64b10097f0c2669ca7cdf4f0d4fad","schema_version":"1.0","event_id":"sha256:ff6558c82b766e2b5cad3f2135eba9e7aea64b10097f0c2669ca7cdf4f0d4fad"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:BPPUG7ZVE4WJBRJIH4L5HZVESZ","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Agentic Recommender System with Hierarchical Belief-State Memory","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"A three-tier belief-state memory with LLM-scheduled lifecycle operations improves personalized recommendation accuracy.","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Benyu Zhang, Hong Yan, Lei Huang, Lizhu Zhang, Qianqian Zhong, Siyu Lin, Xiangjun Fan, Xiang Shen, Yifan Wu, Yuhang Zhou, Zhuokai Zhao","submitted_at":"2026-05-14T05:38:24Z","abstract_excerpt":"Memory-augmented LLM agents have advanced personalized recommendation, yet existing approaches universally adopt flat memory representations that conflate ephemeral signals with stable preferences, and none provides a complete lifecycle governing how memory should evolve. We propose MARS (Memory-Augmented Agentic Recommender System), a framework that treats recommendation as a partially observable problem and maintains a structured belief state that progressively abstracts noisy behavioral observations into a compact estimate of user preferences. MARS organizes this belief state into three tie"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experiments on four InstructRec benchmark domains show that MARS achieves state-of-the-art performance with average improvements of 26.4% in HR@1 and 10.3% in NDCG@10 over the strongest baselines with further gains from agentic scheduling in evolving settings.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the LLM-based planner can accurately and adaptively manage the memory lifecycle operations to maintain a coherent estimate of user preferences without introducing significant errors or biases.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"MARS uses hierarchical memory and LLM planning to achieve 26.4% higher HR@1 on InstructRec benchmarks compared to prior methods.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A three-tier belief-state memory with LLM-scheduled lifecycle operations improves personalized recommendation accuracy.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"33546cfdfe8a4806bc94ac5cde1fe9335e228e74ed69d6fd8b932d3be6082935"},"source":{"id":"2605.14401","kind":"arxiv","version":1},"verdict":{"id":"aa585864-7dc4-4c07-b8f6-d9e36266fce5","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T02:14:46.423229Z","strongest_claim":"Experiments on four InstructRec benchmark domains show that MARS achieves state-of-the-art performance with average improvements of 26.4% in HR@1 and 10.3% in NDCG@10 over the strongest baselines with further gains from agentic scheduling in evolving settings.","one_line_summary":"MARS uses hierarchical memory and LLM planning to achieve 26.4% higher HR@1 on InstructRec benchmarks compared to prior methods.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the LLM-based planner can accurately and adaptively manage the memory lifecycle operations to maintain a coherent estimate of user preferences without introducing significant errors or biases.","pith_extraction_headline":"A three-tier belief-state memory with LLM-scheduled lifecycle operations improves personalized recommendation accuracy."},"references":{"count":27,"sample":[{"doi":"","year":null,"title":"Bradley Knox, and Smitha Milli","work_id":"b6b132b3-7990-4eaa-a937-b6fab400fdc2","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Recusersim: A realistic and diverse user simulator for evaluating conversational recommender systems","work_id":"f9276fe9-7b65-48f1-8c5d-2e22719604ba","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"MemRec: Collaborative Memory-Augmented Agentic Recommender System","work_id":"fb73c690-000a-4578-a45a-622b90b60240","ref_index":3,"cited_arxiv_id":"2601.08816","is_internal_anchor":true},{"doi":"","year":null,"title":"Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory","work_id":"a5aed26c-a248-48b6-a59e-f7693fcb180a","ref_index":4,"cited_arxiv_id":"2504.19413","is_internal_anchor":true},{"doi":"","year":null,"title":"Interactive recommendation agent with active user commands","work_id":"98c34f3a-b079-42f2-a1a3-40eb3c8deac4","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":27,"snapshot_sha256":"29cc6dd6deeb4c5d6ada0ccfcd9e02931562deba9b2122d2988573575d3379e5","internal_anchors":7},"formal_canon":{"evidence_count":2,"snapshot_sha256":"d3a8d8b38a903f31a4121c9b2466830109ab1246b407e756581509f94a8c811b"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"aa585864-7dc4-4c07-b8f6-d9e36266fce5"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:07Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"JqJ3kvFSmlf85pD0TY5EE2HdSp1CKqYLdf4EWf+lBqqpHi+BWIWnM2Dcxd59S9cjssAp0aUsYtJY599fsDXOBQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-28T23:47:42.348672Z"},"content_sha256":"bf9892da9f065c877cfb2026d8a9558ba77809ae650814be6868d671c3ddfd93","schema_version":"1.0","event_id":"sha256:bf9892da9f065c877cfb2026d8a9558ba77809ae650814be6868d671c3ddfd93"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/BPPUG7ZVE4WJBRJIH4L5HZVESZ/bundle.json","state_url":"https://pith.science/pith/BPPUG7ZVE4WJBRJIH4L5HZVESZ/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/BPPUG7ZVE4WJBRJIH4L5HZVESZ/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-05-28T23:47:42Z","links":{"resolver":"https://pith.science/pith/BPPUG7ZVE4WJBRJIH4L5HZVESZ","bundle":"https://pith.science/pith/BPPUG7ZVE4WJBRJIH4L5HZVESZ/bundle.json","state":"https://pith.science/pith/BPPUG7ZVE4WJBRJIH4L5HZVESZ/state.json","well_known_bundle":"https://pith.science/.well-known/pith/BPPUG7ZVE4WJBRJIH4L5HZVESZ/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:BPPUG7ZVE4WJBRJIH4L5HZVESZ","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"156daef7a2fed65ad764a16f6963bfa8b54637036ce4bdf6d0d17da8de80a9a3","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-14T05:38:24Z","title_canon_sha256":"e86ef8f2ae665fde5c754939c76dc6ce142ff8832a7d5fd7690fa085862edf3e"},"schema_version":"1.0","source":{"id":"2605.14401","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.14401","created_at":"2026-05-17T23:39:07Z"},{"alias_kind":"arxiv_version","alias_value":"2605.14401v1","created_at":"2026-05-17T23:39:07Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14401","created_at":"2026-05-17T23:39:07Z"},{"alias_kind":"pith_short_12","alias_value":"BPPUG7ZVE4WJ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"BPPUG7ZVE4WJBRJI","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"BPPUG7ZV","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:bf9892da9f065c877cfb2026d8a9558ba77809ae650814be6868d671c3ddfd93","target":"graph","created_at":"2026-05-17T23:39:07Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Experiments on four InstructRec benchmark domains show that MARS achieves state-of-the-art performance with average improvements of 26.4% in HR@1 and 10.3% in NDCG@10 over the strongest baselines with further gains from agentic scheduling in evolving settings."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the LLM-based planner can accurately and adaptively manage the memory lifecycle operations to maintain a coherent estimate of user preferences without introducing significant errors or biases."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"MARS uses hierarchical memory and LLM planning to achieve 26.4% higher HR@1 on InstructRec benchmarks compared to prior methods."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"A three-tier belief-state memory with LLM-scheduled lifecycle operations improves personalized recommendation accuracy."}],"snapshot_sha256":"33546cfdfe8a4806bc94ac5cde1fe9335e228e74ed69d6fd8b932d3be6082935"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"d3a8d8b38a903f31a4121c9b2466830109ab1246b407e756581509f94a8c811b"},"paper":{"abstract_excerpt":"Memory-augmented LLM agents have advanced personalized recommendation, yet existing approaches universally adopt flat memory representations that conflate ephemeral signals with stable preferences, and none provides a complete lifecycle governing how memory should evolve. We propose MARS (Memory-Augmented Agentic Recommender System), a framework that treats recommendation as a partially observable problem and maintains a structured belief state that progressively abstracts noisy behavioral observations into a compact estimate of user preferences. MARS organizes this belief state into three tie","authors_text":"Benyu Zhang, Hong Yan, Lei Huang, Lizhu Zhang, Qianqian Zhong, Siyu Lin, Xiangjun Fan, Xiang Shen, Yifan Wu, Yuhang Zhou, Zhuokai Zhao","cross_cats":["cs.AI"],"headline":"A three-tier belief-state memory with LLM-scheduled lifecycle operations improves personalized recommendation accuracy.","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-14T05:38:24Z","title":"Agentic Recommender System with Hierarchical Belief-State Memory"},"references":{"count":27,"internal_anchors":7,"resolved_work":27,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Bradley Knox, and Smitha Milli","work_id":"b6b132b3-7990-4eaa-a937-b6fab400fdc2","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Recusersim: A realistic and diverse user simulator for evaluating conversational recommender systems","work_id":"f9276fe9-7b65-48f1-8c5d-2e22719604ba","year":2025},{"cited_arxiv_id":"2601.08816","doi":"","is_internal_anchor":true,"ref_index":3,"title":"MemRec: Collaborative Memory-Augmented Agentic Recommender System","work_id":"fb73c690-000a-4578-a45a-622b90b60240","year":null},{"cited_arxiv_id":"2504.19413","doi":"","is_internal_anchor":true,"ref_index":4,"title":"Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory","work_id":"a5aed26c-a248-48b6-a59e-f7693fcb180a","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Interactive recommendation agent with active user commands","work_id":"98c34f3a-b079-42f2-a1a3-40eb3c8deac4","year":null}],"snapshot_sha256":"29cc6dd6deeb4c5d6ada0ccfcd9e02931562deba9b2122d2988573575d3379e5"},"source":{"id":"2605.14401","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-15T02:14:46.423229Z","id":"aa585864-7dc4-4c07-b8f6-d9e36266fce5","model_set":{"reader":"grok-4.3"},"one_line_summary":"MARS uses hierarchical memory and LLM planning to achieve 26.4% higher HR@1 on InstructRec benchmarks compared to prior methods.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"A three-tier belief-state memory with LLM-scheduled lifecycle operations improves personalized recommendation accuracy.","strongest_claim":"Experiments on four InstructRec benchmark domains show that MARS achieves state-of-the-art performance with average improvements of 26.4% in HR@1 and 10.3% in NDCG@10 over the strongest baselines with further gains from agentic scheduling in evolving settings.","weakest_assumption":"That the LLM-based planner can accurately and adaptively manage the memory lifecycle operations to maintain a coherent estimate of user preferences without introducing significant errors or biases."}},"verdict_id":"aa585864-7dc4-4c07-b8f6-d9e36266fce5"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:ff6558c82b766e2b5cad3f2135eba9e7aea64b10097f0c2669ca7cdf4f0d4fad","target":"record","created_at":"2026-05-17T23:39:07Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"156daef7a2fed65ad764a16f6963bfa8b54637036ce4bdf6d0d17da8de80a9a3","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-14T05:38:24Z","title_canon_sha256":"e86ef8f2ae665fde5c754939c76dc6ce142ff8832a7d5fd7690fa085862edf3e"},"schema_version":"1.0","source":{"id":"2605.14401","kind":"arxiv","version":1}},"canonical_sha256":"0bdf437f35272c90c5283f17d3e6a49651a6f971ad8c64623f01634207375faf","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"0bdf437f35272c90c5283f17d3e6a49651a6f971ad8c64623f01634207375faf","first_computed_at":"2026-05-17T23:39:07.481733Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:39:07.481733Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"OIBFxrFQWvlBOXaWb1YFh4zpoEpr0jA/Dd0n/f946mybH5V39EUdrfDGsK8tlVWdZ8n7miF6np7ehzVUl1KDCg==","signature_status":"signed_v1","signed_at":"2026-05-17T23:39:07.482474Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.14401","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:ff6558c82b766e2b5cad3f2135eba9e7aea64b10097f0c2669ca7cdf4f0d4fad","sha256:bf9892da9f065c877cfb2026d8a9558ba77809ae650814be6868d671c3ddfd93"],"state_sha256":"cd5a5845b1091360caff3ecbf013df3ffedf21de983e4587c572ed60b45e510a"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"VjtVIIdM44ASWjUvLweGpHKkowDSghxTq33gU+bbeLftyPrJNQljNsuR/ranyh3AhjS0cOsQKc6vLVtNW5r1Aw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-28T23:47:42.352552Z","bundle_sha256":"66021304ff17dd181c135bafd6593a1e274d13da48351bb237c4c253f381e0fe"}}