[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-84180-en":3,"doc-seo-84180-105":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},84180,1374391974468,"Eden","https://ap-avatar.wpscdn.com/davatar_29158cc5080c5b710cf443261637dec0",8,"Research & Report","Seeing and Reflecting: Multimodal Memory-Enhanced Agent Collaboration for Recommendation","Large language model (LLM)-based agentic recommender systems can reason over natural-language interactions, yet they are limited by text-centric inputs and coarse memory updates, which lead to missed visual evidence, semantic noise, and preference drift. MMEACR presents a Multimodal Memory-Enhanced Agent Collaboration framework for recommendation. It uses dual-track memory: reasoning agents with attribute-guided reinforcement-and-reflection, and a multimodal embedding memory for fine-grained cross-modal signals. Weighted Reciprocal Rank Fusion integrates both tracks for robust, interpretable rankings.","Seeing and Reflecting: Multimodal Memory-Enhanced Agent Collaboration for Recommendation  \nHao Cong1,* , Huizu Lin2,* , Zihan Wang3,* , Chengkai Huang4,5,\\# ,  \nQuan Z. Sheng5 , Lina Yao4,6  \n1Tsinghua University, 2University of Science and Technology of China, 3Peking University, 4The University of New South Wales,  \n5Macquarie University, 6 CSIRO’s Data61  \n*Equal contribution. \\#Corresponding author.  \nCorrespondence: [chengkai.huang1@unsw.edu.au](chengkai.huang1@unsw.edu.au)  \narXiv :2607 .07 108v 1 [ cs .IR] 8 Jul 2026  \nAbstract  \nLarge language model (LLM)-based agentic recommender systems show promise in modeling user preferences through natural-language reasoning, yet they remain limited by textcentric inputs and coarse-grained memory updates, making agents prone to missing visual evidence, semantic noise, and preference drift. To address these limitations, we propose MMEACR, a Multimodal MemoryEnhanced Agent Collaboration framework for recommendation. MMEACR introducesa dual-track memory architecture that separates interpretable agent reasoning from finegrained multimodal matching. In the reasoning track, collaborative User and Item Memory Agents maintain persistent multimodal memories and update them through an attributeguided reinforcement-and-reflection mechanism. In the matching track, a decoupled multimodal embedding memory is built from raw interaction narratives and item images to preserve detailed cross-modal signals beyond structured memory updates. The two tracks are integrated through weighted Reciprocal Rank Fusion to produce robust and interpretable rankings. Experiments on three real-world domains show that MMEACR achieves strong overall performance against competitive LLM-based and agent-based baselines, with notable gains in visually grounded recommendation scenarios.  \n1 Introduction  \nLarge language models (LLMs) have recently enabled a new class of agentic recommender systems that model user preferences through naturallanguage reasoning and interaction (Huang et al., 2025b, 2024a, 2025a) . Representative methods such as AgentCF (Zhang et al., 2024) formulate recommendation as a collaborative process among language agents, where agents compare candidate  \n(a) Traditional  \nagent-based (b) MMEACR framework  \nrecommendation  \nFigure 1: Comparison between conventional textcentric agent-based recommendation and MMEACR. MMEACR combines multimodal agent memory evolution with embedding-based cross-modal matching and fuses their rankings via Reciprocal Rank Fusion (RRF) .  \nitems, generate rationales, and update their internal memories. This paradigm improves interpretability and flexibility by externalizing preference modeling into language-mediated reasoning rather than relying solely on gradient-based representation learning.  \nDespite their promise, existing LLM-based recommendation agents still suffer from two key limitations. First, most methods remain text-centric, constructing user and item memories mainly from textual histories, reviews, or metadata while underusing visual evidence such as product images (Liu et al., 2025) . This restricts their ability to model preferences in visually grounded domains such as fashion, electronics, and media products, where appearance, style, packaging, and other visual attributes often play an important role. Second, cur-  \nrent memory update mechanisms are often coarsegrained. Agents typically revise their profiles through free-form reflection or direct history accumulation, which can introduce redundant descriptions, amplify spurious rationales, and cause preference drift over repeated interactions. These issues become more challenging when textual and visual signals must be jointly interpreted and selectively consolidated.  \nTo address these challenges, we propose MMEACR, a Multimodal Memory-Enhanced Agent Collaboration framework for recommendation. MMEACR is motivated by the idea of Seeing and Reflecting. Seeing grounds agents in multimodal evidence by initial","cbCairUQ7eO5LxPn","https://ap.wps.com/l/cbCairUQ7eO5LxPn","pdf",1857546,1,16,"English","en",105,"# Abstract\n# Introduction\n## Motivation and Limitations\n## Proposed Framework: Seeing and Reflecting\n### Reasoning Track: Attribute-Guided Memory Evolution\n### Matching Track: Multimodal Embedding Memory\n## Ranking Integration and Contributions","[{\"question\":\"What limitations do existing LLM-based recommendation agents face?\",\"answer\":\"They often rely on text-centric histories and coarse-grained memory updates. This underuses visual evidence and can cause semantic noise and preference drift during repeated interactions.\"},{\"question\":\"How does MMEACR incorporate visual information?\",\"answer\":\"MMEACR initializes and enriches item memories using both textual metadata and image-derived descriptions. It also maintains raw multimodal interaction narratives and item images for embedding-based matching.\"},{\"question\":\"How are the two recommendation tracks combined in MMEACR?\",\"answer\":\"MMEACR produces one ranking from the reasoning track and another from the embedding-based matching track, then fuses them using weighted Reciprocal Rank Fusion for robust and interpretable results.\"}]",1784193693,40,{"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},"seeing-and-reflecting-multimodal-memory-enhanced-agent-collaboration-for-recommendation","",{"@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/seeing-and-reflecting-multimodal-memory-enhanced-agent-collaboration-for-recommendation/84180/",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 limitations do existing LLM-based recommendation agents face?","Question",{"text":75,"@type":76},"They often rely on text-centric histories and coarse-grained memory updates. This underuses visual evidence and can cause semantic noise and preference drift during repeated interactions.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does MMEACR incorporate visual information?",{"text":80,"@type":76},"MMEACR initializes and enriches item memories using both textual metadata and image-derived descriptions. It also maintains raw multimodal interaction narratives and item images for embedding-based matching.",{"name":82,"@type":73,"acceptedAnswer":83},"How are the two recommendation tracks combined in MMEACR?",{"text":84,"@type":76},"MMEACR produces one ranking from the reasoning track and another from the embedding-based matching track, then fuses them using weighted Reciprocal Rank Fusion for robust and interpretable results.","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},[93,97,101,105,110,115,119,122,127,130,134],{"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":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":28,"slug":118},7,"Healthcare","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"]