[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85777-en":3,"doc-seo-85777-105":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},85777,2336464648322,"Aria","https://ap-avatar.wpscdn.com/avatar/2200025388227c56fec?_k=1778556882303663488",8,"Research & Report","An LLM-powered Agentic Recommendation System for Connected TV Content Discovery","Recommendation systems for Connected TV (CTV) face difficulty integrating heterogeneous contextual signals—such as trending topics, breaking news, cultural events, and cross-surface user activity—into ranking pipelines that were originally built for structured behavioral signals. An LLM-powered agentic approach is proposed to reason over and synthesize diverse inputs across varying schemas naturally. Where current LLM-only methods fall short on retrieval efficiency, personalization precision, and scalability, a hybrid agent architecture selects the best method per sub-task, emphasizing practical success and reduced inference latency.","arXiv :2607 .09988v 1 [ cs .IR] 10 Jul 2026  \nAn LLM-powered Agentic Recommendation System for Connected TV Content Discovery  \nLei Shi 1 , Di Wang 1 , Harry Tran 1 , Helsing Xu 1 , Yuchen Lu 1 , Dhara Ghodasara 1 , Wilson Chaney 1 , Xueting Liao 1 , Jerry Yu 1 , Huayu Ding 1 , Mingze Gao 1 , Shike Mei 1 , Shuo Tang 1 , Zhe Zhang 1 , Jianming He 1 , Abhishek Kumar 1 , Haotian Wu 1 , Hamed Firooz 1 , Li Li 1  \n1 Meta  \nRecommendation systems, from traditional multi-stage to recent unified generative architectures, face challenges in incorporating diverse contextual signals, such as trending topics, breaking news, cultural events, and cross-surface user activities, into their ranking pipelines. These systems are designed to consume structured behavioral signals with consistent schemas, and lack the reasoning capability to naturally process unstructured or heterogeneously formatted contextual information. Incorporating such signals typically requires feature engineering, bespoke data pipelines, and carefully tuned heuristics. In this paper, we present an LLM-powered agentic recommendation system designed for Connected TV (CTV) content discovery that addresses these limitations. Our system leverages the reasoning capabilities of large language models to naturally process and synthesize diverse signals across varying schemas and structures, eliminating much of the manual integration inherent in traditional ranking and retrieval systems. Recognizing that current LLM-based solutions still fall short of traditional machine learning models in several recommendation tasks, including retrieval efficiency, personalization precision, and scalability, we adopt an agentic architecture that orchestrates specialized components, allowing each sub-task to be handled by the most suitable method, whether LLM-based or traditional ML. The main contribution of this work is our engineering approach to successfully overcoming the practical limitations of enabling LLM for recommendation, particularly inference latency. We share insights from our work and discuss the trade-offs and lessons learned in building a hybrid system that combines the flexibility of LLMs with the performance of established recommendation techniques.  \nDate: July 14, 2026  \nCorrespondence: Lei Shi at [leis@meta.com](leis@meta.com), Li Li at [leeley@meta.com](leeley@meta.com)  \n1 Introduction  \nConnected TV (CTV) has emerged as a growing surface for content consumption, offering a lean-back, large-screen experience that is distinct from mobile-centric environments. While conventional recommendation interfaces present users with a single ranked list of content, such as infinite-scroll news feeds or short-form video queues, the CTV home screen shows multiple horizontal carousels organized around coherent topics or themes. This architectural shift encourages users to explore content across diverse categories. Consequently, this paradigm transitions the recommendation objective from producing a single total ordering of items to curating multiple, semantically distinct content groupings. These groupings must maintain strict intra-group coherence while collectively maximizing coverage across both evergreen user interests and real-time trends.  \nFurthermore, the lean-back nature of television viewing shapes user expectations distinctly from mobile consumption. On CTV surfaces, users expect curated, timely, and contextually rich content, analogous to a traditional television programming guide. Topics tied to current events, trending cultural moments, breaking news, local happenings, and seasonal themes carry heightened importance on CTV. This contrasts sharply with mobile surfaces, where algorithmic personalization based primarily on historical engagement signals dominates. Meeting these expectations requires recommendation systems capable of incorporating real-world context to surface content that feels editorially curated yet remains personalized.  \nExisting recommendation systems, howev","cbCailuSzoJbwfAg","https://ap.wps.com/l/cbCailuSzoJbwfAg","pdf",795125,1,13,"English","en",105,"# Introduction\n## Connected TV recommendation objectives and UI paradigm\n## Context-aware requirements versus mobile personalization\n## Limitations of existing recommendation pipelines\n## Related generative and unified recommendation architectures\n## Motivation for an agentic hybrid design","[{\"question\":\"What problem does the agentic recommendation system target for Connected TV content discovery?\",\"answer\":\"It targets the challenge of incorporating diverse, real-world contextual signals—like trending topics and breaking news—into recommendation ranking pipelines that mainly handle structured behavioral signals.\"},{\"question\":\"How does the system use LLM reasoning to address heterogeneous contextual information?\",\"answer\":\"It leverages large language models to naturally process and synthesize signals across different schemas and structures, reducing the manual integration typically needed in traditional retrieval and ranking setups.\"},{\"question\":\"Why is a hybrid agentic architecture used instead of relying only on LLM-based recommendation?\",\"answer\":\"The work notes that current LLM-based solutions can lag behind traditional machine learning on tasks such as retrieval efficiency, personalization precision, and scalability, so sub-tasks are orchestrated to use either LLM-based or traditional ML methods as appropriate.\"}]",1784206208,33,{"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},"an-llm-powered-agentic-recommendation-system-for-connected-tv-content-discovery","",{"@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/an-llm-powered-agentic-recommendation-system-for-connected-tv-content-discovery/85777/",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},"What problem does the agentic recommendation system target for Connected TV content discovery?","Question",{"text":75,"@type":76},"It targets the challenge of incorporating diverse, real-world contextual signals—like trending topics and breaking news—into recommendation ranking pipelines that mainly handle structured behavioral signals.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does the system use LLM reasoning to address heterogeneous contextual information?",{"text":80,"@type":76},"It leverages large language models to naturally process and synthesize signals across different schemas and structures, reducing the manual integration typically needed in traditional retrieval and ranking setups.",{"name":82,"@type":73,"acceptedAnswer":83},"Why is a hybrid agentic architecture used instead of relying only on LLM-based recommendation?",{"text":84,"@type":76},"The work notes that current LLM-based solutions can lag behind traditional machine learning on tasks such as retrieval efficiency, personalization precision, and scalability, so sub-tasks are orchestrated to use either LLM-based or traditional ML methods as appropriate.","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},[93,97,101,105,110,115,120,123,128,131,135],{"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":111,"doc_module":4,"doc_module_name":45,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":45,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":45,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":45,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":45,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]