recsys-search-ads skillA
recsys-search-ads is agent-read markdown (skill) from yuecao365/offercome: 搜推广深挖：召回与向量检索、精排多目标、重排多样性、样本与偏差、冷启动探索、搜索相关性、广告校准、生成式推荐。.
Indexed from public GitHub and served as immutable, content-addressed versions. Install it pinned to an exact SHA-256 with the mdr CLI, and every file is verified against the hash recorded here before it reaches your agent. The deterministic audit below grades the latest version, and the same file always earns the same grade.
What the file says
## 面试官在意什么 搜索推荐广告是国内算法岗需求最大的方向，面试有一套自己的语言：召回粗排精排重排的候选量级、双塔的样本构造、多目标怎么融合、曝光偏差怎么纠、冷启动怎么做、离线 AUC 涨了线上没涨怎么办。2026 年多了一层：生成式推荐（用大模型或序列生成替代部分排序）与 LLM 做特征和标注在哪些场景真的有用。面试官追的是链路每一层的数字与一次线上没涨的归因。 ## 项目 / 实习怎么深挖 简历上出现下面这类经历时从哪里切、追什么。追到候选人能说出机制、数字的来源与一次真实的故障或取舍才算实；只有框架名与结论、说不出自己那一段的，记为危险信号。通用的追问方法见 project-deep-dive。 - 简历出现推荐 / 广告模型 → 追召回样本构造、负采样、各层候选量级与延迟预算、离线与线上指标的对应、AB 置信度 - 简历出现双塔 / 向量召回 → 追样本对怎么构、batch 内负采样、流行度修正、索引更新频率 - 简历出现多目标 → 追目标之间怎么权衡、融合公式怎么定、线上怎么调 - 简历出现搜索相关性 / 排序 → 追标注怎么来、离线 NDCG 与线上点击的关系、query 理解与召回的分工 - 简历出现 CTR / CVR 预估 → 追校准怎么做、延迟反馈怎么处理、出价与预估偏差怎么互相影响 - 简历出现冷启动 / 探索 → 追新用户新物品各怎么处理、探索流量多少、效果怎么量 - 简历出现生成式推荐 / LLM 用于搜推 → 追替代了哪一层、延迟与成本、和传统模型对比的实测 ## 常见失守与危险信号 - 推荐系统链路：说不出各层的候选量级与延迟预算；负样本只答"随机采"；不知道曝光未点击样本的偏差 - 搜索与广告排序：相关性只答"用 BERT"；不知道 query 理解与召回的分工；广告只谈 CTR 不谈出价与校准 - 样本构造与不平衡：直接把负样本采到 1:1 且不做校准；不知道延迟反馈 - 召回与向量检索：只会说"用双塔"；不知道索引类型与更新代价；没有流行度修正 - 精排与多目标：目标融合拍脑袋；不知道 MMoE / PLE 解决什么；序列建模只会报名字 - 重排与多样性：不知道重排存在；多样性靠规则硬打散 - 冷启动与探索：只答"用规则"；探索流量没有量化 - 生成式推荐与 LLM：无条件用 LLM；或完全拒绝而说不出理由；没有延迟与成本账 ## 常考主题清单 只列名字、阶梯与答实的标志，作"问到哪一层算实"的参考；问哪些、问几道由这份 JD 与这份简历定，不是配额。 ### 样本构造与不平衡 - 阶梯：正负样本怎么定义 → 采样、加权、focal loss 各解决什么 → 负采样后概率校准偏了怎么修 → 标签噪声与延迟反馈（转化在点击后几天才发生）怎么处理 …
Read the whole file at its exact version.
How to install
mdr add yuecao365/offercome/recsys-search-ads@git:20260920.a4843f5mdr add yuecao365/offercome/recsys-search-ads@sha256:b51a5b5fc8ab9af4Pin to a label to follow the author's releases, or to a sha256 to freeze the exact bytes forever. Either way the resolved hash is written to mdr.lock, and mdr install reproduces it on any machine.
[](https://markdownregistry.com/a/art_qkmmahjnbd3atyqz)
1 badge views in 30 days
Versions
Audit of the latest version
- pass: Frontmatter block present
- pass: Frontmatter declares a name
- pass: Frontmatter declares a description
- pass: Size between 200 bytes and 200 KB (6884 bytes)
- pass: No zero-width or bidi control characters
- pass: No instruction hidden inside an HTML comment
- pass: No link to an exfiltration or paste host
- pass: No credential-shaped string
- pass: No instruction to send local credentials anywhere
- pass: No text hidden with inline styles
- pass: No prompt-injection phrasing
- pass: No curl or wget piped into a shell
- pass: No recursive delete of root, home or parent
- pass: No instruction to read or print local credentials
- pass: No base64 blob over 200 characters
- pass: No link to a raw IP address
- pass: No script tag
Source
yuecao365/offercome · 23 stars · license MIT · pushed 2026-09-23 · branch main
API
GET https://markdownregistry.com/api/v1/artifacts/art_qkmmahjnbd3atyqz GET https://markdownregistry.com/api/v1/resolve?ref=yuecao365/offercome/recsys-search-ads GET https://markdownregistry.com/api/v1/blob/b51a5b5fc8ab9af4637c5f70262e5f881913be30c0d6b917516acf0a54703aa0
Your agent does the legwork. You hear about the deals worth your word. Hand yours the standing instructions at modelranch.com and it joins the network that reads files like this one.