2026-06-26 周五 · Generative

独立报告页 · 在站内浏览(含往期切换)→
Generative生成式推荐 · 2 篇Personalization个性化推荐 · 3 篇AI SearchAI 搜索 · 8 篇Agentic智能体搜索 · 6 篇Infra系统与索引 · 8 篇

今日场景社区搜索 (UGC 短文本搜索): 用户在社区平台搜索帖子、问答、评论等 UGC 内容。核心指标是搜索相关性 (nDCG) 和点击率 (CTR)。关注:语义匹配、query 理解、短文本召回、排序模型、LLM 在搜索中的应用。 — 评分 12 篇,入选 2 篇,分数 3.00 – 4.10

本期目录 · 2 篇
  1. 01Recommendation as Generation: Unifying Personalized Video Generation and Recommendation at Industrial Scale
  2. 02Evaluating Scene-based In-Situ Item Labeling for Immersive Conversational Recommendation
LEAD STORY · 今日必读 · Kuaishou / Beihang

Recommendation as Generation: Unifying Personalized Video Generation and Recommendation at Industrial Scale

快手与北航提出 Recommendation-as-Generation(RaG),把短视频推荐从“在固定池里检索”改为“按推断的用户兴趣实时生成个性化视频”。核心是解耦语义 ID(D-SIDs):用 Qwen2.5-VL-7B 做指令引导的多模态表征,再用 RQ-KMeans 残差量化成 content 与 creative 两套离散 token;视频生成交给共享 Qwen2.5-32B 的三个角色 Agent,并用 GDPO 做跨域奖励学习。快手 4 亿日活广告场景上线 A/B,相对 DLRM 广告收入 +5.462%,码本碰撞率从 18.24% 降到 2.62%。

相关 4新颖 5实用 3证据 4时效 5
4.10
综合分 / 5
偏好调整 4.30
KEY INSIGHT · 对业务意味着什么语义 ID 不必是单一码本:把内容语义与风格语义拆成两套 SID,碰撞率从 18.24% 降到 2.62%,兴趣建模与生成可控性同时受益,这一解耦量化思路对纯推荐的 SID 设计也可直接借鉴。SCRL 用 GDPO 把稀疏真实反馈设为主目标、把视频质量与兴趣对齐设为约束,给“多奖励尺度不一致”提供了可复用写法。代价是 VGA 为延迟瓶颈、目前只能 nearline 服务;资源有限的团队可先用 D-SIDs,暂不碰端到端视频生成。
Yanhua Cheng, Bo Wang, Haotian Zhang, et al. · 2026-06-25 · arXiv
独立页面阅读 ↗

今日其余入选

1 篇 · 按综合排名
02 3.00

Evaluating Scene-based In-Situ Item Labeling for Immersive Conversational Recommendation

UofT

多伦多大学把对话式推荐(CRS)扩展到 XR 沉浸场景,提出 Immersive CRS:推荐项直接高亮在用户视野中并叠加“就地标签(in-situ label)”。论文把信息需求分为“显式意图满足”和“主动信息需求”两类,据此定义标签选择的评测指标,在 fashion/movie/retail 三个场景上对比 IR、LLM、VLM 三类方法,指出现有方法未利用场景特定模态、给出视觉可推断的冗余信息、且仅凭对话难以预判主动信息需求。

INSIGHT评测/问题定义型论文,对社区搜索的价值在评测方法:把“呈现什么信息”与“推荐什么物品”拆开评估,并区分显式意图与主动需求,可迁移到搜索结果摘要/卡片的信息取舍。它实测出 LLM/VLM 在场景模态利用与冗余抑制上的系统性不足,提醒做对话式/生成式搜索别只靠对话文本。局限是聚焦 XR 沉浸交互、与 UGC 文本搜索差距较大,借鉴点更多在评测范式层面。

Jiazhou Liang, Yifan Simon Liu, David Guo, Minqi Sun, Yilun Jiang, Scott Sanner
arXiv·2026-04-06·Relevance 2 · Novelty 3 · Practical 3 · Evidence 4 · Timely 5

其他赛道今日

点击切换
Personalization
3
个性化推荐 · 篇入选
AI Search
8
AI 搜索 · 篇入选
Agentic
6
智能体搜索 · 篇入选
Infra
8
系统与索引 · 篇入选

开源雷达

GitHub / HuggingFace 热门 · 8 个项目
llm-value-rankings: Daily LLM value rankings - compare 300+ models by intelligence, speed and price. OpenRouter + Artificial Analysis. 大模型性价比排行榜Daily LLM value rankings - compare 300+ models by intelligence, speed and price. OpenRouter + Artificial Analysis. 大模型性价比排行榜JavaScript · ★ 5 · 2026-06-23 · GitHubclaudecode-glm-stack: Run GLM-5.2 inside Claude Code through Z.ai with isolated settingsRun GLM-5.2 inside Claude Code through Z.ai with isolated settings★ 5 · 2026-06-21 · GitHubFinAgents: Multi-agent LLM-powered financial analysis framework for stocks.Multi-agent LLM-powered financial analysis framework for stocks.Python · ★ 14 · 2026-06-21 · GitHubQwen/Qwen-AgentWorld-35B-A3Bpipeline: text-generation; tags: transformers, safetensors, qwen3_5_moe, image-text-to-text, qwen, world-model, agent, environment-simulation★ 0 · 2026-06-22 · HuggingFaceQwen/AgentWorldBenchAgentWorldBench AgentWorldBench is a comprehensive evaluation benchmark for language world models, constructed from real-world observations of frontier model trajectories on established benchmarks such as Tool Decathlon, Terminal-Bench 1.0 & 2.0, and OSWorld-Verified. Every evalu; tags: task_categories:text-generation, language:en, license:apache-2.0, size_categories:1K<n<10K, format:json, modality:tabular, modality:text, library:datasets★ 0 · 2026-06-22 · HuggingFaceinternlm/RNGBench-Game-TrajectoriesRNGBench-Game-Trajectories SFT (supervised fine-tuning) trajectory data accompanying RNGBench · Reconstructive Non-Markov Games — an evaluation framework that tests whether multimodal language models can reconstruct hidden state from memory and act on it in closed-loop environmen; tags: task_categories:image-text-to-text, language:en, license:mit, size_categories:10K<n<100K, format:json, modality:text, library:datasets, library:pandas★ 0 · 2026-06-22 · HuggingFacedeepseek-ai/DeepSeek-V4-Propipeline: text-generation; tags: transformers, safetensors, deepseek_v4, text-generation, conversational, arxiv:2606.19348, license:mit, eval-results★ 0 · 2026-04-22 · HuggingFacezai-org/GLM-5.2pipeline: text-generation; tags: transformers, safetensors, glm_moe_dsa, text-generation, conversational, en, zh, arxiv:2602.15763★ 0 · 2026-06-16 · HuggingFace