{"id":"https://openalex.org/W7167608771","doi":"https://doi.org/10.48550/arxiv.2607.02818","title":"Session-Level Optimization for Large-Scale Retrieval using REINFORCE with Multi-Step Off-Policy Correction","display_name":"Session-Level Optimization for Large-Scale Retrieval using REINFORCE with Multi-Step Off-Policy Correction","publication_year":2026,"publication_date":"2026-07-02","ids":{"openalex":"https://openalex.org/W7167608771","doi":"https://doi.org/10.48550/arxiv.2607.02818"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2607.02818","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2607.02818","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2607.02818","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5022052407","display_name":"Artem Matveev","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Matveev, Artem","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5068390397","display_name":"Sergei Makeev","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Makeev, Sergei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5140213591","display_name":"Aleksei Krasilnikov","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Krasilnikov, Aleksei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5031658527","display_name":"Vladimir Baikalov","orcid":"https://orcid.org/0009-0009-4864-2305"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Baikalov, Vladimir","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5119556708","display_name":"Sergei Liamaev","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liamaev, Sergei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5091814356","display_name":"Kirill Khrylchenko","orcid":"https://orcid.org/0009-0007-3640-8795"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Khrylchenko, Kirill","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10203","display_name":"Recommender Systems and Techniques","score":0.6877999901771545,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10203","display_name":"Recommender Systems and Techniques","score":0.6877999901771545,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10286","display_name":"Information Retrieval and Search Behavior","score":0.15690000355243683,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.03310000151395798,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/autoregressive-model","display_name":"Autoregressive model","score":0.609499990940094},{"id":"https://openalex.org/keywords/generative-grammar","display_name":"Generative grammar","score":0.6011999845504761},{"id":"https://openalex.org/keywords/session","display_name":"Session (web analytics)","score":0.5953999757766724},{"id":"https://openalex.org/keywords/generative-model","display_name":"Generative model","score":0.46540001034736633},{"id":"https://openalex.org/keywords/scaling","display_name":"Scaling","score":0.2904999852180481}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7839000225067139},{"id":"https://openalex.org/C159877910","wikidata":"https://www.wikidata.org/wiki/Q2202883","display_name":"Autoregressive model","level":2,"score":0.609499990940094},{"id":"https://openalex.org/C39890363","wikidata":"https://www.wikidata.org/wiki/Q36108","display_name":"Generative grammar","level":2,"score":0.6011999845504761},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.6003999710083008},{"id":"https://openalex.org/C2779182362","wikidata":"https://www.wikidata.org/wiki/Q17126187","display_name":"Session (web analytics)","level":2,"score":0.5953999757766724},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5898000001907349},{"id":"https://openalex.org/C167966045","wikidata":"https://www.wikidata.org/wiki/Q5532625","display_name":"Generative model","level":3,"score":0.46540001034736633},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.30300000309944153},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.2904999852180481},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.26489999890327454},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.26440000534057617},{"id":"https://openalex.org/C557471498","wikidata":"https://www.wikidata.org/wiki/Q554950","display_name":"Recommender system","level":2,"score":0.2556999921798706}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2607.02818","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2607.02818","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2607.02818","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2607.02818","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[{"score":0.7930250763893127,"display_name":"Peace, Justice and strong institutions","id":"https://metadata.un.org/sdg/16"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Two-tower":[0],"models":[1],"are":[2,14],"a":[3,40,47,68,85,109,119],"widely":[4],"used":[5,62],"paradigm":[6],"for":[7,104],"large-scale":[8,142],"retrieval":[9,169],"in":[10,63],"recommendation.":[11],"However,":[12],"they":[13],"typically":[15],"trained":[16],"with":[17,51,132],"myopic":[18],"supervised":[19],"objectives,":[20],"such":[21],"as":[22,39],"next-item":[23,159],"prediction,":[24],"that":[25,89,111,124,146],"do":[26],"not":[27],"directly":[28],"optimize":[29],"long-term":[30,136,183],"user":[31,86,91],"satisfaction.":[32],"In":[33],"this":[34],"work,":[35,65],"we":[36,66,83],"formulate":[37],"recommendation":[38,131],"session-level":[41],"sequential":[42,105],"decision-making":[43,106],"problem":[44],"and":[45,128,160],"train":[46,84],"two-tower":[48],"retriever":[49],"autoregressively":[50],"off-policy":[52,60,102,152],"REINFORCE":[53],"on":[54,139,167],"pre-collected":[55],"data.":[56],"Unlike":[57],"the":[58,76,130,133,140,187],"one-step":[59],"correction":[61],"prior":[64],"propose":[67],"multi-step":[69],"approximation":[70],"of":[71,154],"importance":[72],"weights":[73],"enabled":[74],"by":[75],"autoregressive":[77],"formulation.":[78],"To":[79],"support":[80],"offline":[81],"evaluation,":[82],"feedback":[87],"model":[88],"simulates":[90,125],"responses":[92,127,179],"to":[93,107,176],"generated":[94],"recommendations.":[95],"This":[96],"lets":[97],"us":[98],"adapt":[99],"doubly":[100],"robust":[101],"evaluation":[103],"recommendation,":[108],"setting":[110],"has":[112],"received":[113],"limited":[114],"attention.":[115],"We":[116],"further":[117],"introduce":[118],"feedback-model-based":[120],"test-time":[121],"scaling":[122],"procedure":[123],"future":[126,178],"selects":[129],"highest":[134],"predicted":[135],"return.":[137],"Experiments":[138],"public":[141],"Yambda-5B":[143],"dataset":[144],"show":[145],"our":[147],"RL":[148],"agent":[149],"achieves":[150],"higher":[151,181],"estimates":[153],"cumulative":[155],"session":[156],"reward":[157],"than":[158],"next-positive":[161],"prediction":[162],"baselines,":[163],"while":[164],"remaining":[165],"competitive":[166],"conventional":[168],"metrics.":[170],"Moreover,":[171],"allocating":[172],"more":[173],"inference-time":[174],"compute":[175],"simulating":[177],"yields":[180],"model-based":[182],"returns":[184],"without":[185],"updating":[186],"policy.":[188]},"counts_by_year":[],"updated_date":"2026-07-29T09:40:50.615796","created_date":"2026-07-08T00:00:00"}