2026-08-17 周一 · Personalization

独立报告页 · 在站内浏览(含往期切换)→
Generative生成式推荐 · 3 篇Personalization个性化推荐 · 5 篇AI SearchAI 搜索 · 5 篇Agentic智能体搜索 · 5 篇Infra系统与索引 · 9 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 14 篇,入选 5 篇,分数 2.70 – 4.30

本期目录 · 5 篇
  1. 01FEDIN: Frequency-Enhanced Deep Interest Network for Click-Through Rate Prediction
  2. 02Context-aware diffusion model for sequential recommendation
  3. 03TF-Rec: Time–Frequency Enhanced Sequential Recommendation Based on Adaptive Intent Clustering
  4. 04Gradient-enhanced evolutionary multi-objective optimization (GEEMOO): Balancing relevance, learning outcomes, and diversity in educational recommendation systems
  5. 05Metric Distortion with Preference Intensities.
LEAD STORY · 今日必读

FEDIN: Frequency-Enhanced Deep Interest Network for Click-Through Rate Prediction

论文针对CTR预估中用户兴趣的时序变化与频率模式难以同时利用的问题,提出频率增强Deep Interest Network(FEDIN),在时间域和频率域联合编码行为序列,并用频域分支提取周期性兴趣。在Tmall、Alipay、Taobao上,FEDIN的GAUC/AUC分别为0.9658/0.9666、0.9335/0.9320、0.9740/0.9729,均超过DIN、DIEN等七类基线。

相关 4新颖 4实用 4证据 5时效 4
4.30
综合分 / 5
偏好调整 4.30
KEY INSIGHT · 对业务意味着什么频域建模为用户兴趣提供了“周期信号”视角,适合补足只依赖最近行为的序列模型。工程上可先把频域分支放入精排或预估实验,用GAUC和分桶校准验证它是否真正改善长周期兴趣;论文结果覆盖三个电商数据集,但在线延迟与频谱计算成本仍需单独压测。
Zenan Dai et al. · 2026-07-20 · 网页来源
独立页面阅读 ↗

今日其余入选

4 篇 · 按综合排名
02 4.15

Context-aware diffusion model for sequential recommendation

📖 有深读 · 10 分钟

CADiff面向扩散式序列推荐中长期兴趣表示不完整、长短期兴趣权重固定的问题,提出非对称感知编码器APE与自适应多兴趣融合AMF:前者让序列上下文注入用户特定表示,后者按当前场景动态生成扩散引导信号。模型在MovieLens、KuaiRec、Yelp和Book四个真实数据集上同时改善推荐准确率与个性化效果,并提供了公开代码。

INSIGHT长短期兴趣不应使用全局固定权重,用户当前会话与内容类型决定两者的相对贡献。对社区UGC推荐可借鉴“用户状态+当前序列”的门控融合,再把扩散生成限制在候选集合内;同时要警惕多步采样带来的线上延迟,优先验证蒸馏或少步采样方案。

Hao Chen, Youcai Wang, Yue Xu, Yanyan Chen
网页来源·2026-07-01·Relevance 4 · Novelty 4 · Practical 4 · Evidence 4 · Timely 4
独立页面阅读 ↗
03 3.85

TF-Rec: Time–Frequency Enhanced Sequential Recommendation Based on Adaptive Intent Clustering

TF-Rec针对序列推荐中的多意图与周期性行为,组合三部分机制:用自适应可学习原型聚类发现潜在意图,用意图引导扩散生成刻画动态偏好演化,再以时间-频率多任务对比学习对齐不同视角的行为表示。论文发表于2026年7月19日,核心目标是让兴趣表示同时保留短期变化、周期规律和意图结构。

INSIGHT当UGC用户同时存在“长期主题偏好”和“本次会话意图”时,先聚类意图再做序列建模比单一兴趣向量更可解释。可将原型聚类作为离线用户状态特征接入现有CTR模型,先做轻量A/B;扩散生成与多任务对比学习会增加训练复杂度,应以增量收益和资源成本共同评估。

网页来源·2026-07-19·Relevance 4 · Novelty 4 · Practical 3 · Evidence 4 · Timely 4
04 3.70

Gradient-enhanced evolutionary multi-objective optimization (GEEMOO): Balancing relevance, learning outcomes, and diversity in educational recommendation systems

GEEMOO面向教育推荐中相关性、学习效果和多样性相互冲突的问题,把梯度法的快速整合能力与进化策略的全局搜索结合,寻找Pareto最优解。实验中Hypervolume为0.85、Generational Distance为0.02、Spread Indicator为0.88、Crowding Distance为0.92;相较NSGA-II,适应度评估从60,000次降至50,000次。

INSIGHT多目标推荐不应把CTR、深度消费和内容生态简单线性加权后长期固定,Pareto前沿可以保留多种业务取舍。社区搜索/推荐可借鉴其“候选解集+在线策略选择”思路,将相关性、点击和多样性约束分层验证;但教育场景与UGC场景不同,不能直接迁移指标权重。

Y. Jdidou, S. Aammou, Hicham Er-radi, Ilias Aarab
Semantic Scholar·2025-09-01·Relevance 4 · Novelty 4 · Practical 4 · Evidence 4 · Timely 1
05 2.70

Metric Distortion with Preference Intensities.

论文研究带偏好强度的排序投票:在传统“先后顺序”之外,允许用户表达强烈或普通偏好,并提出Positional Scoring Matching规则,通过零和博弈求出该规则类中的最优成员。在度量失真框架下,利用偏好强度可把失真控制在3以下,论文同时给出了忽略强度信息的代价界。

INSIGHT它不是推荐系统论文,但提醒我们点击顺序并不等于偏好强度:一次快速点击和持续深度消费可能代表不同的满意度。对多目标排序可把停留、收藏、负反馈等信号转成强度或置信度,再做鲁棒聚合;落地前必须处理反馈噪声、可比性和隐私约束。

Mehrad Abbaszadeh, Ali Ansarifar, Mohamad Latifian, Masoud Seddighin
Conference·2026-01-01·Relevance 2 · Novelty 4 · Practical 2 · Evidence 4 · Timely 2

其他赛道今日

点击切换
Generative
3
生成式推荐 · 篇入选
AI Search
5
AI 搜索 · 篇入选
Agentic
5
智能体搜索 · 篇入选
Infra
9
系统与索引 · 篇入选

开源雷达

GitHub / HuggingFace 热门 · 8 个项目
XingGraph: Knowledge-graph RAG for LLM agents: ontology-driven graph building (RDF/OWL), structured-doc chunking with title attribution, multi-subject WIKI-completion retrieval, model-hop traversal, traceable answers on Neo4j. Built for traceable, low-hallucination LLM memory.Knowledge-graph RAG for LLM agents: ontology-driven graph building (RDF/OWL), structured-doc chunking with title attribution, multi-subject WIKI-completion retrieval, model-hop traversal, traceable answers on Neo4j. Built for traceable, low-hallucination LLM memory.Python · ★ 10 · 2026-08-10 · GitHubDragon-Brain: Dragon Brain — persistent long-term memory for AI agents via MCP (Model Context Protocol). Knowledge graph (FalkorDB) + vector search (Qdrant) + CUDA GPU embeddings. Works with Claude, Gemini CLI, Cursor, Windsurf, VS Code Copilot. 30 tools, 1121 tests.Dragon Brain — persistent long-term memory for AI agents via MCP (Model Context Protocol). Knowledge graph (FalkorDB) + vector search (Qdrant) + CUDA GPU embeddings. Works with Claude, Gemini CLI, Cursor, Windsurf, VS Code Copilot. 30 tools, 1121 tests.Python · ★ 50 · 2026-02-23 · GitHubpm-knowledge-qa: Local AI Personal Knowledge Base Q&A · Full‑Link RAG Practical Implementation (Hybrid Retrieval + Rerank Fine‑Ranking + Citation‑Enabled Generation)Local AI Personal Knowledge Base Q&A · Full‑Link RAG Practical Implementation (Hybrid Retrieval + Rerank Fine‑Ranking + Citation‑Enabled Generation)Python · ★ 7 · 2026-08-15 · GitHubinternlm/WildClawBench-TrajectoriesWildClawBench Trajectories Complete OpenClaw agent trajectories from the WildClawBench evaluation — every message, reasoning block, tool call, and tool result from real long-horizon agent runs, released for independent verification, side-by-side comparison, and trace-level analys; tags: task_categories:text-generation, language:en, language:zh, license:mit, size_categories:n<1K, format:parquet, modality:text, library:datasetsdataset · ★ 5 · 2026-07-30 · HuggingFaceinternlm/WildClawBenchWildClawBench Hard, practical, end-to-end evaluation for AI agents — in the wild. WildClawBench is an agent benchmark that tests what actually matters: can an AI agent do real work, end-to-end, without hand-holding? We drop agents into a live OpenClaw environment — the same open-; tags: benchmark:official, benchmark:eval-yaml, task_categories:visual-question-answering, task_categories:image-text-to-text, task_categories:question-answering, language:en, language:zh, license:mitdataset · ★ 68 · 2026-03-24 · HuggingFaceinternlm/WildClawBench-HarborWildClawBench-Harbor This repository is the WildClawBench benchmark converted to the Harbor task format, so that all 60 tasks can be run directly with harbor run against any Harbor-supported agent (Claude Code, OpenHands, Codex CLI, custom agents, ...). WildClawBench is a benchma; tags: language:en, language:zh, license:mit, size_categories:n<1K, arxiv:2605.10912, region:us, agent, agent-evaluationdataset · ★ 2 · 2026-08-12 · HuggingFaceQwen/Qwen3.8-27Bpipeline: image-text-to-text; tags: transformers, safetensors, qwen3_5, image-text-to-text, conversational, license:apache-2.0, eval-results, endpoints_compatiblemodel · ★ 10231 · 2026-08-05 · HuggingFacezai-org/GLM-5pipeline: text-generation; tags: transformers, safetensors, glm_moe_dsa, text-generation, conversational, en, zh, arxiv:2602.15763model · ★ 2122 · 2026-02-11 · HuggingFace