知识卡片

问题思考卡 · 7014.e · 生成模型:DeepSeek-V4.1-Flash

判断该不该用 ML 的五条特征

理解等级:中级

众多 ML 应用虽然存在,但 ML 的核心强项是做决策或预测;作者给出面向人的 ML 系统在生产中通常具备的一项或多项特征,可作为「该不该用 ML」的判据。复杂逻辑:人力定义规则不切实际,如搜索引擎串联多阶段排序(文本索引初检、文本相似度主排序、上下文排序、个性化排序)。大规模个性化:问题预期扩展到数千用户以上,例如 yarnit.ai 预计三到六个月内数千顾客会用上新优惠,需支持大用户群的个性化。规则随时间快速变化:规则若常年不变,启发式更好;若业务成功依赖快速适配与改规则(如用户不断写商品评论、推荐算法需实时调整),ML 更合适。有清晰的评估指标:例如希望推荐能带来成交,搜索 kids hat 应返回与儿童配色、帽子针号相关的花样与毛线。不要求 100% 准确:若高概率正确即可达成业务成功,ML 是好选择——推荐系统不因用户偶尔不想要所展示的东西就算故障,用户体验仍可很好,模型还能从缺乏成交中学习、改进未来推荐。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • How Product and ML Interact · 查看原书依据

      In general, our style of human-facing ML system will have one or more of the following characteristics as deployed in production: Complex logic that’s impractical to solve with human-defined rules For example, search engines often have multiple phases of ranking that happen in series, such as initial retrieval from the text index, primary ranking using text similarity, contextual ranking, and personalized ranking. Large-scale

    参考来源

    - 位置:《Reliable Machine Learning》第12章《How Product and ML Interact》(源文件:extracted/Reliable Machine Learning/_epub-src/OEBPS/ch12.xhtml) - 结论依据:原文明确阐述该主题,并给出定义、实践或反例。 - 原始内容:In general, our style of human-facing ML system will have one or more of the following characteristics as deployed in production: Complex logic that’s impractical to solve with human-defined rules For example, search engines often have multiple phases of ranking that happen in series, such as initial retrieval from the text index, primary ranking using text similarity, contextual ranking, and personalized ranking. Large-scale