{"paper":{"title":"SimPersona: Learning Discrete Buyer Personas from Raw Clickstreams for Grounded E-Commerce Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"SimPersona learns discrete buyer types from clickstreams to let LLM agents simulate diverse real buyer populations in e-commerce.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Alberto Castelo, Han Li, Lingyun Wang, Shuang Xie, Ted Chaiwachirasak, Zahra Zanjani Foumani","submitted_at":"2026-05-14T00:01:11Z","abstract_excerpt":"LLM-based web agents can navigate live storefronts, yet they often collapse to a single \"average buyer\" policy, failing to capture the heterogeneous and distributional nature of real buyer populations. Existing personalization methods rely on hand-crafted prompt-based personas that are brittle, difficult to scale, context-inefficient, and unable to faithfully represent population-level behavior. We introduce SimPersona, a novel framework that learns discrete buyer types from historical traffic and exposes them to LLM-based web agents as compact persona tokens. Given raw clickstreams, a behavio"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Evaluated on 8.37M buyers across 42 held-out live storefronts, SimPersona achieves 78% conversion-rate alignment with real buyers, exhibits interpretable behavioral variation across buyer types, and outperforms a baseline with 8× more parameters on goal-oriented shopping tasks.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The discrete buyer types learned from historical clickstreams will transfer effectively to LLM agent behavior in new live interactions without major distribution shift or loss of fidelity.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SimPersona uses VQ-VAE to induce discrete buyer types from clickstreams, maps them to LLM persona tokens, and fine-tunes agents to achieve 78% conversion-rate alignment with real buyers across 42 storefronts.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"SimPersona learns discrete buyer types from clickstreams to let LLM agents simulate diverse real buyer populations in e-commerce.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"6f93ca10aae8a5fb546895a22ac5058ad7b46ddda69fc8f530bf60f25ff45f24"},"source":{"id":"2605.14205","kind":"arxiv","version":1},"verdict":{"id":"5e6cb387-d2e2-478c-888c-176bc54fb474","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T02:52:59.932218Z","strongest_claim":"Evaluated on 8.37M buyers across 42 held-out live storefronts, SimPersona achieves 78% conversion-rate alignment with real buyers, exhibits interpretable behavioral variation across buyer types, and outperforms a baseline with 8× more parameters on goal-oriented shopping tasks.","one_line_summary":"SimPersona uses VQ-VAE to induce discrete buyer types from clickstreams, maps them to LLM persona tokens, and fine-tunes agents to achieve 78% conversion-rate alignment with real buyers across 42 storefronts.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The discrete buyer types learned from historical clickstreams will transfer effectively to LLM agent behavior in new live interactions without major distribution shift or loss of fidelity.","pith_extraction_headline":"SimPersona learns discrete buyer types from clickstreams to let LLM agents simulate diverse real buyer populations in e-commerce."},"references":{"count":35,"sample":[{"doi":"","year":2007,"title":"k-means++: The advantages of careful seeding","work_id":"ffbfe61b-cb77-4b64-a921-fa4713e96fee","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1974,"title":"A dendrite method for cluster analysis.Communications in Statistics – Theory and Methods, 3(1):1–27, 1974","work_id":"bfbe7651-0e14-4401-9191-be17bc96c872","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Beyond demographics: Aligning role-playing llm-based agents using human belief networks","work_id":"4dade001-b288-440f-b9cd-3a9c10e96919","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1988,"title":"Lawrence Erlbaum Associates, 2 edition","work_id":"d4f908c1-d99f-4777-b721-828c5e0d4b1b","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Mind2web: Towards a generalist agent for the web","work_id":"45afb2f7-1446-4115-a5fb-463dbdd1a2e0","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":35,"snapshot_sha256":"d81f7cc9d98c22e7140e88f6ddd82c964a319e3991ec09f9b1d6a8e7f1e17087","internal_anchors":6},"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"}