[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85755-en":3,"doc-seo-85755-105":28,"detail-sidebar-cat-0-en-105":90},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":11,"language":21,"language_code":22,"site_id":23,"html_lang":22,"table_of_contents":24,"faqs":25,"seo_title":13,"seo_description":14,"update_tm":26,"read_time":27},85755,5909877438554,"Maeve","https://ap-avatar.wpscdn.com/avatar/5600025385ad2bf12a7?_k=1778553567797529272",8,"Research & Report","RouteRec: 严格评估推荐代理的选择与聚合","Recommender systems face a heterogeneous-agents choice where collaborative filters, sequential models, content retrievers, and LLM rerankers each have different costs and strengths. The study introduces RouteRec, a task-aware agent-ranking framework under cost constraints, comparing request-level hard selection versus item-level learned aggregation across four traditional agents and one LLM reranker. On MovieLens-1M, the oracle shows strong headroom (HR@10=0.584), yet leakage-free 5-fold evaluation shows hard selection underperforms BM25, while item-level learned aggregation improves quality and uses LLM calls selectively.","RouteRec: Strict Evaluation of Recommender-Agent Selection  \nand Aggregation  \nKaiji Zhou  \nUniversity of Birmingham Birmingham, United Kingdom [kxz571@student.bham.ac.uk](kxz571@student.bham.ac.uk)  \nVladimir Kalmykov  \nUniversity of Birmingham Birmingham, United Kingdom [vxk276@student.bham.ac.uk](vxk276@student.bham.ac.uk)  \nYue Feng∗ University of Birmingham Birmingham, United Kingdom [y.feng.6@bham.ac.uk](y.feng.6@bham.ac.uk)  \narXiv :2607 .09908v 1 [ cs .CL] 10 Jul 2026  \nAbstract  \nRecommender systems increasingly face a choice among heterogeneous agents—collaborative filters, sequential models, contentbased retrievers, and LLM-based rerankers—yet no single agent is uniformly best. We study this choice as task-aware agent ranking under cost constraints using RouteRec, a framework that compares request-level hard selection with item-level learned aggregation over four traditional recommender agents and one LLM reranker agent. On MovieLens-1M, the full quality oracle has substantial headroom (HR@10=0.584), confirming that useful cross-agent signal exists. Under a leakage-free 5-fold out-of-fold protocol, however, hard selection remains below BM25 (0 . 223 vs. 0. 254), and selective LLM escalation does not improve it. The same protocol yields a different outcome for learned aggregation: its cheap-only variant matches BM25 in HR and has a higher NDCG point estimate (0.123 vs. 0.114), while gated all-agent aggregation reaches HR@10= 0.295 with 70.2% LLM calls. The resulting lesson is not that routing is solved, but that request-level selection of one complete agent list is too coarse for this sparse fixed-candidate setting; item-level aggregation is the more promising action space.  \n1 Introduction  \nRecent work has explored LLMs as recommendation agents—as zero-shot rankers [9], instruction followers [30], and tool users [31] . However, LLM inference can be substantially more expensive than conventional recommendation inference [3, 18], and LLMs do not uniformly dominate cheaper alternatives. Traditional recommendation methods remain competitive for warm users with rich interaction histories [13], while surveys discuss potential LLM advantagesin sparse, cross-domain, and explanation-oriented settings [14, 28] .  \nThis observation motivates a shift from “which single model is best?” to “which model should handle this request?”—a framing consistent with classical algorithm selection [21], meta-learning for recommendation [4, 6, 23], and LLM routing [18, 25] .  \nWe use recommender agent as an operational umbrella term for any callable recommendation service with a ranked-list interface, a cost profile, and task-dependent suitability. Accordingly, our pool contains four cheap traditional recommender agents and one expensive LLM reranker agent. This definition matches the agent-search setting, where systems must discover, compare, and invoke heterogeneous callable services; it does not require every service to be an autonomous dialogue agent.  \nWe study this routing problem through RouteRec, a lightweight framework for per-instance decisions across the five recommender agents. RouteRec-Select combines request context with cheap probe  \n∗ Corresponding author.  \nFigure 1: RouteRec-Select architecture. A request is encoded and combined with cheap-probe disagreement features to select the best cheap agent (Stage 1) and decide whether to escalate to the LLM reranker (Stage 2).  \ndisagreement, a deployable measure of cheap-agent output divergence, to choose one cheap agent and decide whether to escalate to the LLM (Fig. 1) . Because hard selection returns one complete list from one agent, it can discard useful evidence from other agents. We therefore also evaluate RouteRec-Stack, which reranks the union of agent top-􀀺 lists using deployable item-level rank and score evidence.  \nThis paper makes three contributions. First, we cast search over recommender agents as task-aware utility maximisation under cost constraints and evaluate eve","cbCaipG0q9JjzK4A","https://ap.wps.com/l/cbCaipG0q9JjzK4A","pdf",898753,1,"English","en",105,"# Abstract\n# Introduction\n## Motivation for request-level routing\n## RouteRec framework and decision stages\n# Related Work\n## LLMs for recommendation\n## Hybrid recommenders and algorithm selection","[{\"question\":\"What problem does RouteRec address in recommender systems?\",\"answer\":\"RouteRec targets task-aware routing among heterogeneous recommendation agents, deciding which agent(s) should handle a given request under strict cost constraints.\"},{\"question\":\"How does RouteRec compare request-level selection with item-level aggregation?\",\"answer\":\"RouteRec-Select performs request context–based hard selection and optional LLM escalation, while RouteRec-Stack aggregates signals at the item level by reranking unions of agents’ candidate lists.\"},{\"question\":\"What do results on MovieLens-1M show about hard selection versus aggregation?\",\"answer\":\"Under a leakage-free 5-fold out-of-fold protocol, hard selection stays below BM25, while learned aggregation—especially the gated all-agent variant—recovers more cross-agent signal and achieves higher HR@10 with selective LLM calls.\"}]",1784206036,20,{"code":4,"msg":29,"data":30},"ok",{"site_id":23,"language":22,"slug":31,"title":13,"keywords":32,"description":14,"schema_data":33,"social_meta":85,"head_meta":87,"extra_data":89,"updated_unix":26},"routerec-strict-evaluation-of-recommender-agent-selection-and-aggregation","",{"@graph":34,"@context":84},[35,52,67],{"@type":36,"itemListElement":37},"BreadcrumbList",[38,42,46,49],{"item":39,"name":40,"@type":41,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":43,"name":44,"@type":41,"position":45},"https://docshare.wps.com/document/","Document",2,{"item":47,"name":12,"@type":41,"position":48},"https://docshare.wps.com/document/research-report/",3,{"item":50,"name":13,"@type":41,"position":51},"https://docshare.wps.com/document/routerec-strict-evaluation-of-recommender-agent-selection-and-aggregation/85755/",4,{"url":50,"name":13,"@type":53,"author":54,"headline":13,"publisher":56,"fileFormat":59,"inLanguage":22,"description":14,"dateModified":60,"datePublished":61,"encodingFormat":59,"isAccessibleForFree":62,"interactionStatistic":63},"DigitalDocument",{"name":9,"@type":55},"Person",{"url":39,"name":57,"@type":58},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":64,"interactionType":65,"userInteractionCount":20},"InteractionCounter",{"@type":66},"ViewAction",{"@type":68,"mainEntity":69},"FAQPage",[70,76,80],{"name":71,"@type":72,"acceptedAnswer":73},"What problem does RouteRec address in recommender systems?","Question",{"text":74,"@type":75},"RouteRec targets task-aware routing among heterogeneous recommendation agents, deciding which agent(s) should handle a given request under strict cost constraints.","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"How does RouteRec compare request-level selection with item-level aggregation?",{"text":79,"@type":75},"RouteRec-Select performs request context–based hard selection and optional LLM escalation, while RouteRec-Stack aggregates signals at the item level by reranking unions of agents’ candidate lists.",{"name":81,"@type":72,"acceptedAnswer":82},"What do results on MovieLens-1M show about hard selection versus aggregation?",{"text":83,"@type":75},"Under a leakage-free 5-fold out-of-fold protocol, hard selection stays below BM25, while learned aggregation—especially the gated all-agent variant—recovers more cross-agent signal and achieves higher HR@10 with selective LLM calls.","https://schema.org",{"og:url":50,"og:type":86,"og:title":13,"og:site_name":57,"og:description":14},"article",{"robots":88,"canonical":50},"index,follow",{"doc_id":7,"site_id":23},{"code":4,"msg":5,"data":91},[92,96,100,104,109,114,119,122,126,129,133],{"id":20,"doc_module":4,"doc_module_name":44,"category_name":93,"show_sort_weight":94,"slug":95},"Story & Novel",90,"story-novel",{"id":45,"doc_module":4,"doc_module_name":44,"category_name":97,"show_sort_weight":98,"slug":99},"Literature",80,"literature",{"id":51,"doc_module":4,"doc_module_name":44,"category_name":101,"show_sort_weight":102,"slug":103},"Exam",70,"exam",{"id":105,"doc_module":4,"doc_module_name":44,"category_name":106,"show_sort_weight":107,"slug":108},5,"Comic",60,"comic",{"id":110,"doc_module":4,"doc_module_name":44,"category_name":111,"show_sort_weight":112,"slug":113},6,"Technology",50,"technology",{"id":115,"doc_module":4,"doc_module_name":44,"category_name":116,"show_sort_weight":117,"slug":118},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":44,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":44,"category_name":124,"show_sort_weight":27,"slug":125},9,"Religion & Spirituality","religion-spirituality",{"id":27,"doc_module":4,"doc_module_name":44,"category_name":127,"show_sort_weight":27,"slug":128},"World Cup","world-cup",{"id":130,"doc_module":4,"doc_module_name":44,"category_name":131,"show_sort_weight":130,"slug":132},10,"Lifestyle","lifestyle",{"id":134,"doc_module":4,"doc_module_name":44,"category_name":135,"show_sort_weight":105,"slug":136},19,"General","general"]