{"paper":{"title":"PersonalAlign: Hierarchical Implicit Intent Alignment for Personalized GUI Agent with Long-Term User-Centric Records","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Hierarchical memory from long-term records lets GUI agents resolve vague instructions and anticipate routines.","cross_cats":["cs.CV","cs.HC","cs.LG"],"primary_cat":"cs.AI","authors_text":"Gongwei Chen, Liqiang Nie, Rui Shao, Weili Guan, Yibo Lyu","submitted_at":"2026-01-14T17:12:48Z","abstract_excerpt":"While GUI agents have shown strong performance under explicit and completion instructions, real-world deployment requires aligning with users' more complex implicit intents. In this work, we highlight Hierarchical Implicit Intent Alignment for Personalized GUI Agent (PersonalAlign), a new agent task that requires agents to leverage long-term user records as persistent context to resolve omitted preferences in vague instructions and anticipate latent routines by user state for proactive assistance. To facilitate this study, we introduce AndroidIntent, a benchmark designed to evaluate agents' ab"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"HIM-Agent significantly improves both execution and proactive performance by 15.7% and 7.3% on the AndroidIntent benchmark.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the 20k long-term records contain stable, generalizable user preferences and routines that transfer to new vague instructions without overfitting to the 20 users in the benchmark.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"PersonalAlign introduces a hierarchical memory agent that uses long-term user records to resolve vague GUI instructions and provide proactive assistance, improving execution by 15.7% and proactive performance by 7.3% on the new AndroidIntent benchmark.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Hierarchical memory from long-term records lets GUI agents resolve vague instructions and anticipate routines.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"bfeff95e40675015a848a096ab8edb634897c4bfb28916f3c8ac4c2a2bca5da2"},"source":{"id":"2601.09636","kind":"arxiv","version":2},"verdict":{"id":"d044c2cf-fa1f-4681-8c24-c470b471c708","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T14:29:14.352576Z","strongest_claim":"HIM-Agent significantly improves both execution and proactive performance by 15.7% and 7.3% on the AndroidIntent benchmark.","one_line_summary":"PersonalAlign introduces a hierarchical memory agent that uses long-term user records to resolve vague GUI instructions and provide proactive assistance, improving execution by 15.7% and proactive performance by 7.3% on the new AndroidIntent benchmark.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the 20k long-term records contain stable, generalizable user preferences and routines that transfer to new vague instructions without overfitting to the 20 users in the benchmark.","pith_extraction_headline":"Hierarchical memory from long-term records lets GUI agents resolve vague instructions and anticipate routines."},"references":{"count":14,"sample":[{"doi":"","year":2025,"title":"LLaVA-OneVision-1.5: Fully Open Framework for Democratized Multimodal Training","work_id":"41c2802e-aff9-482f-b506-10955ff0838d","ref_index":1,"cited_arxiv_id":"2509.23661","is_internal_anchor":true},{"doi":"","year":2025,"title":"Score the steps, not just the goal: Vlm-based subgoal evaluation for robotic manipulation","work_id":"ebb97240-cc8c-40cc-8684-46c3b7df07da","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"InProceedings of the IEEE/CVF conference on com- puter vision and pattern recognition, pages 10023– 10031","work_id":"cdfa2461-8017-4be7-b0a6-ad555737130d","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Advances in Neural Information Processing Systems, 37:52040–52094","work_id":"ba5bc0df-5aa8-497c-9121-7b4ac95aa9bb","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"MemAgent: Reshaping Long-Context LLM with Multi-Conv RL-based Memory Agent","work_id":"ac66fe72-b869-4874-8a9d-c110b849da0c","ref_index":5,"cited_arxiv_id":"2507.02259","is_internal_anchor":true}],"resolved_work":14,"snapshot_sha256":"427899431cf6e8c5308e1ad4f6bc09c0a5db2c0b063264f2b2a6d2c9593edd37","internal_anchors":2},"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"}