今日场景今天这条赛道只有一篇,来自淘宝的拍照搜商品排序。它关心的是一个很常见的做法:排序模型里并排放好几种特征交叉网络,各自训练、最后合并。作者让这些分支在训练时互相纠错,并约束它们学到的中间向量既有关联又不雷同,线上点击率和成交额都有提升。 — 评分 7 篇,入选 1 篇,分数 4.50 – 4.50
LEAD STORY · 今日必读 ·
Alibaba
Multi-Branch Cooperation Networks for Enhanced Click-Through Rate Prediction in Large-Scale E-Commerce Search
做排序的团队值得读,因为它指出了多分支模型一个常被忽略的浪费:各分支擅长的样本不同,训练时却从不互相学。淘宝拍立淘(拍照搜商品)团队让三个分支在训练中互相纠错:同一条样本上,一个分支判对了用户点不点、另一个判错了,就用判对那个的预测去带判错的那个。另加一项约束,要求一个分支的中间向量乘上一个正交矩阵后,接近另一个分支的向量。线上 A/B 中点击率从 10.37% 提到 10.46%,GMV 相对提升 1.62%。多分支的收益不只来自多放几个分支,还来自让它们在彼此出错的样本上互相纠正。
4.50
综合分 / 5
偏好调整 4.55
偏好调整 4.55
Xu Chen, Zida Cheng, Shuai Xiao, Chen Ju, Xiaoming Liu · 2026-05-28 · DOI
其他赛道今日
点击切换开源雷达
GitHub / HuggingFace 热门 · 8 个项目musicurgy/delta-inference-engineDelta Inference Engine: DeepSeek-V4.1-Flash on one NVIDIA GPU, natively on Windows (OpenAI-compatible server)C++ · ★ 47 · 2026-10-26 · GitHubaryabhuptani/nyan-qwen3.5-4bQwen3.5-4B catgirl neuron steering research: reproducible experiments and validationPython · ★ 25 · 2026-10-28 · GitHubkarminski/deepseek-v4.1-flash-1017-issue-povDeepSeek Agent 问题 POV repo.JavaScript · ★ 21 · 2026-10-24 · GitHubAlibaba-DAMO-Academy/RynnUnify-Agentic-Reasonerpipeline: image-text-to-text; tags: transformers, safetensors, qwen3_vl, image-text-to-text, rynnunify, embodied-ai, visual-reasoning, reinforcement-learningmodel · ★ 3 · 2026-10-22 · HuggingFaceAlibaba-DAMO-Academy/Rynn-VPRynn-VP: Long-Horizon Visual Planning Benchmark Rynn-VP evaluates visual plans for everyday manipulation in real indoor scenes. Given a natural-language task and an initial image, a system produces a candidate video or an ordered sequence of step images. Evaluation examines the a; tags: language:en, license:cc-by-nc-4.0, size_categories:n<1K, region:us, visual-planning, embodied-ai, video-generation, multimodaldataset · ★ 1 · 2026-10-25 · HuggingFaceAlibaba-DAMO-Academy/RynnUnify-GEU-BenchRynnUnify-GEU-Bench RynnUnify-GEU-Bench is the GEU (Generation Enhanced Understanding) benchmark used for the RynnUnify family. It contains 2000 multi-turn samples across 10 subsets and 3 task families, each paired with its source images and a stored rollout video. Each record wa; tags: task_categories:visual-question-answering, language:en, license:other, size_categories:1K<n<10K, region:us, embodied-ai, spatial-reasoning, video-understandingdataset · ★ 1 · 2026-10-29 · HuggingFaceTencentARC/WorldCrafter-Fastpipeline: image-to-video; tags: diffusers, safetensors, arxiv:2609.24984, worldcrafter, text-to-video, camera-control, image-to-video, diffusers:WorldCraftermodel · ★ 24 · 2026-09-21 · HuggingFaceinternlm/Atria-Dawntags: safetensors, glm_moe_dsa, arxiv:2609.15818, region:usmodel · ★ 14 · 2026-10-23 · HuggingFace