[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-83629-en":3,"doc-seo-83629-105":29,"detail-sidebar-cat-0-en-105":91},{"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":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},83629,1649267921044,"Ava Thompson","https://us-avatar.wpscdn.com/avatar/1800007509477c92dfb?_k=1782875107921204101",8,"Research & Report","Planning over Matrix Factorization MDPs for Candidate Generation","Recommender services treat a user’s journey as a chain of item choices where each recommendation updates user state and therefore the next candidates. Standard matrix-factorization retrieval scores items independently from a single static user vector, ignoring state dynamics. The work studies when it is beneficial to plan over these dynamics by modeling top-k retrieval as an MDP built on the implicit-ALS posterior, using closed-form fold-in transitions and rewards combining relevance and posterior alignment. Experiments compare static retrieval, one-step planning, and horizon-k MCTS on five datasets under two splitting protocols. Lookahead often captures most gains, improving candidate generation without retraining or changing embeddings, with cosine-based relevance critical to avoid entanglement with popularity.","Planning over Matrix-Factorization MDPs for Candidate Generation  \nMikhail Trapeznikov  \nAI VK Moscow, Russia  \nLomonosov Moscow State University Moscow, Russia [trapeznikovmy@my.msu.ru](trapeznikovmy@my.msu.ru)  \nMaksim Utushkin  \nAI VK Moscow, Russia [mak.utushkin@gmail.com](mak.utushkin@gmail.com)  \narXiv :2607 .02 1 15v 1 [ cs .IR] 2 Jul 2026  \nAbstract  \nFor a recommender service, we view the customer journey as a chain of item recommendations: a useful item changes the user’s state and therefore what should be retrieved next. Standard matrixfactorization retrieval ignores this—it builds one user vector and returns the top-􀀠 items by a static score, treating them as independent. We ask a narrow question: when is it worth planning over the user-state dynamics that fold-in induces? To answer it we propose casting top-􀀠 retrieval as an MDP over the implicit-ALS posterior (􀀖−1,􀁄) , where an action is an item and the transition is a closed-form rank-one fold-in, and the trajectory reward combines a relevance similarity with a posterior-alignment term. Under the same fixed embeddings we compare static retrieval, one-step planning, and horizon-􀀠 MCTS across five datasets and two protocols: a per-user leave-last-􀀽 split and a stricter global time split. Dynamicsaware planning tends to overcome static retrieval on all datasets under leave-last-􀀽, and the gains hold on MovieLens-1M and the VK-LSVD slices under the global time split. A single step of lookahead already captures most of the gain, so the lightweight planning layer turns static top-􀀠 scoring into a short decision and improves retrieval over fixed collaborative-filtering embeddings, with noretraining and no change to the representation. These gains depend on measuring relevance with cosine rather than inner-product similarity, which is otherwise entangled with item popularity.  \nCCS Concepts  \n• Information systems → Recommender systems; • Computing methodologies → Reinforcement learning; Neural networks.  \nKeywords  \nrecommender systems, candidate generation, matrix factorization, reinforcement learning, planning, Monte-Carlo tree search  \nACM Reference Format:  \nMikhail Trapeznikov and Maksim Utushkin. 2026. Planning over MatrixFactorization MDPs for Candidate Generation. In 5th Workshop on Endto-End Customer Journey Optimization at KDD 2026, August 09, 2026, Jeju Island, Republic of Korea. ACM, New York, NY, USA, 6 pages.  \nPermission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than the author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission [and/or a fee. Request permissions from permissions@acm.org](and/or a fee. Request permissions from permissions@acm.org).  \nKDD CJ Workshop ’26, Jeju Island, Republic of Korea  \n© 2026 Copyright held by the owner/author(s) . Publication rights licensed to ACM.  \n1 Introduction  \nRecommender systems help users navigate large catalogs of products, videos, music and other items, and so they shape the customer experience on most digital platforms. A single recommendation is rarely an individual event: a good pick moves the user further along their journey, a poor one ends the session or pushes them into an irrelevant part of the catalog. So a recommendation always comes with two effects – the immediate choice, and the way it changes what the user will value next.  \nModern systems take this seriously by modelling the user history as a sequence of interactions and feeding it through learned sequence encoders – RNNs and, more recently, transformer-based architectures [4, 7, 19]. The underlying view is an MDP [6, 16]: the state describes th","cbCaiaXDF16JfEkR","https://ap.wps.com/l/cbCaiaXDF16JfEkR","pdf",3670415,1,6,"English","en",105,"# Introduction\n## Recommender journeys as MDPs\n## Static matrix-factorization retrieval limitations\n## Decision layers over fixed embeddings","[{\"question\":\"Why is standard matrix-factorization candidate retrieval insufficient for customer journey optimization?\",\"answer\":\"It constructs a single user embedding and ranks top-k items by a static score, treating candidates as independent. This ignores how consuming earlier recommendations changes the implicit-ALS posterior and should affect subsequent choices.\"},{\"question\":\"How does the proposed method formulate top-k retrieval as a decision process?\",\"answer\":\"Top-k retrieval is cast as an MDP over the implicit-ALS posterior, where an action selects an item and the transition follows a closed-form rank-one fold-in. The trajectory reward combines relevance similarity with a posterior-alignment term.\"},{\"question\":\"What approaches are compared in the experiments and what do the results show?\",\"answer\":\"The paper compares static retrieval, one-step planning, and horizon-k MCTS using the same fixed embeddings across multiple datasets and two evaluation splits. Dynamics-aware planning improves over static retrieval, and a single lookahead step captures most of the benefit.\"}]",1784189386,15,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":27},"planning-over-matrix-factorization-mdps-for-candidate-generation","",{"@graph":35,"@context":85},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/planning-over-matrix-factorization-mdps-for-candidate-generation/83629/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"Why is standard matrix-factorization candidate retrieval insufficient for customer journey optimization?","Question",{"text":75,"@type":76},"It constructs a single user embedding and ranks top-k items by a static score, treating candidates as independent. This ignores how consuming earlier recommendations changes the implicit-ALS posterior and should affect subsequent choices.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does the proposed method formulate top-k retrieval as a decision process?",{"text":80,"@type":76},"Top-k retrieval is cast as an MDP over the implicit-ALS posterior, where an action selects an item and the transition follows a closed-form rank-one fold-in. The trajectory reward combines relevance similarity with a posterior-alignment term.",{"name":82,"@type":73,"acceptedAnswer":83},"What approaches are compared in the experiments and what do the results show?",{"text":84,"@type":76},"The paper compares static retrieval, one-step planning, and horizon-k MCTS using the same fixed embeddings across multiple datasets and two evaluation splits. Dynamics-aware planning improves over static retrieval, and a single lookahead step captures most of the benefit.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,114,119,122,127,130,134],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":45,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":21,"doc_module":4,"doc_module_name":45,"category_name":111,"show_sort_weight":112,"slug":113},"Technology",50,"technology",{"id":115,"doc_module":4,"doc_module_name":45,"category_name":116,"show_sort_weight":117,"slug":118},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":45,"category_name":124,"show_sort_weight":125,"slug":126},9,"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":45,"category_name":128,"show_sort_weight":125,"slug":129},"World Cup","world-cup",{"id":131,"doc_module":4,"doc_module_name":45,"category_name":132,"show_sort_weight":131,"slug":133},10,"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":45,"category_name":136,"show_sort_weight":106,"slug":137},19,"General","general"]