information-architecture
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
ChineseInformation Architecture
信息架构（IA）
Purpose
目的
Information architecture (IA) is the practice of organizing, structuring, and labeling content to help users find and manage information effectively. Good IA makes complex information navigable, discoverable, and understandable.
Use this skill when:
- Designing navigation for websites, apps, documentation, or knowledge bases
- Restructuring content that users can't find or understand
- Creating taxonomies for classification, tagging, or metadata
- Organizing information at scale (hundreds or thousands of items)
- Improving findability when search and browse both fail
- Designing mental models that match how users think about content
Information architecture bridges user mental models and system structure. The goal: users can predict where information lives and find it quickly.
信息架构（IA）是对内容进行组织、结构化和标记的实践，旨在帮助用户高效地查找和管理信息。优质的IA能让复杂信息具备可导航性、可发现性和易懂性。
在以下场景使用本技能：
- 设计导航：为网站、应用、文档或知识库搭建导航系统
- 重构内容：优化用户无法找到或理解的内容结构
- 创建分类体系：用于内容分类、标记或元数据管理
- 规模化组织信息：处理成百上千条量级的内容
- 提升可查找性：当搜索和浏览功能均无法满足需求时
- 设计心智模型：构建与用户对内容的认知匹配的模型
信息架构是连接用户心智模型与系统结构的桥梁，核心目标是让用户能够预判信息的位置并快速找到它。
Common Patterns
常见模式
Pattern 1: Content Audit → Card Sort → Sitemap
模式1：内容审计 → 卡片分类 → 站点地图
When: Redesigning existing site/app with lots of content
Process:
- Content audit: Inventory all existing content (URLs, titles, metadata)
- Card sorting: Users group content cards into categories
- Analyze patterns: What categories emerge? What's grouped together?
- Create sitemap: Translate patterns into hierarchical structure
- Validate with tree testing: Can users find content in new structure?
Example: E-commerce site with 500 products. Audit products → Card sort with 15 users → Patterns show users group by "occasion" not "product type" → New navigation: "Daily Essentials", "Special Occasions", "Gifts" instead of "Electronics", "Clothing", "Home Goods"
适用场景：对已有大量内容的网站/应用进行重设计
流程：
- 内容审计：盘点所有现有内容（URL、标题、元数据）
- 卡片分类：让用户将内容卡片分组为不同类别
- 分析模式：提炼用户形成的分类逻辑和内容分组方式
- 创建站点地图：将用户的分类模式转化为层级结构
- 树状测试验证：测试用户能否在新结构中找到目标内容
示例：拥有500个产品的电商网站。先对产品进行审计 → 邀请15位用户进行卡片分类 → 发现用户按“使用场景”而非“产品类型”分组 → 新导航改为「日常必备」「特殊场合」「礼品」，替代原有的「电子产品」「服饰」「家居用品」
Pattern 2: Taxonomy Design (Faceted Navigation)
模式2：分类体系设计（多面导航）
When: Users need multiple ways to slice/filter information
Structure: Orthogonal facets (dimensions) that combine
- Facet 1: Category (e.g., "Shoes", "Shirts", "Pants")
- Facet 2: Brand (e.g., "Nike", "Adidas", "Puma")
- Facet 3: Price range (e.g., "$0-50", "$50-100", "$100+")
- Facet 4: Color, Size, etc.
Principle: Facets are independent. Users can filter by any combination.
Example: Amazon product browse. Filter by Category AND Brand AND Price simultaneously. Each facet narrows results without breaking others.
适用场景：用户需要从多个维度筛选信息时
结构：采用相互独立的筛选维度（面）
- 维度1：品类（如「鞋履」「上衣」「裤子」）
- 维度2：品牌（如「Nike」「Adidas」「Puma」）
- 维度3：价格区间（如「$0-50」「$50-100」「$100+」）
- 维度4：颜色、尺码等
原则：各筛选维度相互独立，用户可任意组合筛选条件
示例：亚马逊商品浏览功能，用户可同时按品类、品牌、价格筛选，每个维度的筛选结果不会影响其他维度的可用性
Pattern 3: Progressive Disclosure (Hub-and-Spoke)
模式3：渐进式披露（枢纽-分支结构）
When: Content hierarchy is deep, users need overview before details
Structure:
- Hub page: High-level overview with clear labels
- Spoke pages: Detailed content, linked from hub
- Breadcrumbs: Show path back to hub
Principle: Don't overwhelm with everything at once. Start simple, reveal complexity on-demand.
Example: Documentation site. Hub: "Getting Started" with 5 clear options (Install, Configure, First App, Tutorials, Troubleshooting). Each option links to detailed spoke. Users scan hub, pick entry point, dive deep, return to hub if stuck.
适用场景：内容层级较深，用户需要先获取概览再查看详情时
结构：
- 枢纽页面：展示清晰标签的高层级概览
- 分支页面：详细内容页面，从枢纽页面跳转进入
- 面包屑导航：显示返回枢纽页面的路径
原则：避免一次性展示过多信息，从简单内容开始，按需展示复杂内容
示例：文档网站。枢纽页面「快速开始」包含5个清晰选项（安装、配置、首个应用、教程、故障排查），每个选项链接到对应的详细分支页面。用户可快速浏览枢纽页面选择入口，深入查看详情，若遇到问题可返回枢纽页面。
Pattern 4: Flat vs. Deep Navigation
模式4：扁平式 vs 深层式导航
When: Deciding navigation depth (breadth vs. depth tradeoff)
Flat navigation (broad, shallow):
- Structure: Many top-level categories, few sub-levels (e.g., 10 categories, 2 levels deep)
- Pros: Less clicking, everything visible
- Cons: Overwhelming choice, hard to scan 10+ options
Deep navigation (narrow, tall):
- Structure: Few top-level categories, many sub-levels (e.g., 5 categories, 5 levels deep)
- Pros: Manageable choices at each level (5-7 items)
- Cons: Many clicks to reach content, users get lost in depth
Optimal: 3-4 levels deep, 5-9 items per level (Hick's Law: more choices = longer decision time)
Example: Software docs. Flat: All 50 API methods visible at once (overwhelming). Deep: APIs → Authentication → Methods → JWT → jwt.sign() (5 clicks, frustrating). Optimal: APIs (8 categories) → Authentication (6 methods) → jwt.sign() (3 clicks).
适用场景：决定导航深度（广度与深度的权衡）
扁平式导航（广而浅）：
- 结构：大量顶级分类，极少子层级（如10个分类，2层深度）
- 优势：减少点击次数，所有内容可见
- 劣势：选项过多易造成选择困难，10+选项难以快速浏览
深层式导航（窄而深）：
- 结构：少量顶级分类，多层子层级（如5个分类，5层深度）
- 优势：每个层级的选项数量可控（5-7个）
- 劣势：到达目标内容需多次点击，用户易在层级中迷失
最优方案：3-4层深度，每层5-9个选项（希克定律：选项越多，决策时间越长）
示例：软件文档。扁平式：一次性展示50个API方法（过于繁杂）；深层式：API → 认证 → 方法 → JWT → jwt.sign()（5次点击，体验糟糕）；最优：API（8个分类）→ 认证（6个方法）→ jwt.sign()（3次点击）
Pattern 5: Mental Model Alignment (Card Sorting)
模式5：心智模型对齐（卡片分类法）
When: You don't know how users think about content
Process:
- Open card sort: Users create their own categories (exploratory)
- Closed card sort: Users fit content into your categories (validation)
- Hybrid card sort: Users use your categories OR create new ones (refinement)
- Analyze: What labels do users use? What groupings emerge? What's confusing?
Example: SaaS product features. Company calls them "Widgets", "Modules", "Components" (technical terms). Card sort reveals users think "Reports", "Dashboards", "Alerts" (task-based terms). Insight: Label by user tasks, not internal architecture.
适用场景：不了解用户对内容的认知逻辑时
流程：
- 开放式卡片分类：用户自行创建分类（探索性研究）
- 封闭式卡片分类：用户将内容归入预设分类（验证性研究）
- 混合式卡片分类：用户可使用预设分类或创建新分类（优化研究）
- 分析结果：用户使用哪些标签？形成了哪些分组？哪些内容易混淆？
示例：SaaS产品功能。公司内部称其为「Widgets」「Modules」「Components」（技术术语），但卡片分类发现用户用「报表」「仪表盘」「告警」（任务导向术语）来归类。洞察：需以用户任务而非内部架构来命名标签。
Pattern 6: Tree Testing (Reverse Card Sort)
模式6：树状测试（反向卡片分类）
When: Validating navigation structure before building
Process:
- Create text-based tree (sitemap without visuals)
- Give users tasks: "Where would you find X?"
- Track paths: What route did they take? Did they succeed?
- Measure: Success rate, directness (fewest clicks), time
Example: Navigation tree with "Services → Web Development → E-commerce". Task: "Find information about building an online store". 80% success = good. 40% success = users don't understand "E-commerce" label or "Services" category. Iterate.
适用场景：在开发前验证导航结构的合理性
流程：
- 创建基于文本的树状结构（无视觉元素的站点地图）
- 给用户分配任务：「你会在哪里找到X内容？」
- 追踪路径：用户选择了哪些路径？是否成功找到？
- 衡量指标：成功率、直接性（最少点击次数）、耗时
示例：导航树为「服务 → 网页开发 → 电商」。任务：「找到搭建在线商店的相关信息」。80%成功率=结构良好；40%成功率=用户不理解「电商」标签或「服务」分类，需迭代优化。
Workflow
工作流程
Use this structured approach when designing or auditing information architecture:
□ Step 1: Understand context and users
□ Step 2: Audit existing content (if any)
□ Step 3: Conduct user research (card sorting, interviews)
□ Step 4: Design taxonomy and navigation
□ Step 5: Create sitemap and wireframes
□ Step 6: Validate with tree testing
□ Step 7: Implement and iterate
□ Step 8: Monitor findability metricsStep 1: Understand context and users (details)
Identify content volume, user goals, mental models, and success metrics (time to find, search queries, bounce rate).
Step 2: Audit existing content (details)
Inventory all content (URLs, titles, metadata). Identify duplicates, gaps, outdated items. Measure current performance (analytics, heatmaps).
Step 3: Conduct user research (details)
Run card sorting (open, closed, or hybrid) with 15-30 users. Analyze clustering patterns, category labels, outliers. Conduct user interviews to understand mental models.
Step 4: Design taxonomy and navigation (details)
Create hierarchical structure (3-4 levels, 5-9 items per level). Design facets for filtering. Choose labeling system (task-based, audience-based, or alphabetical). Define metadata schema.
Step 5: Create sitemap and wireframes (details)
Document structure visually (sitemap diagram). Create low-fidelity wireframes showing navigation, breadcrumbs, filters. Get stakeholder feedback.
Step 6: Validate with tree testing (details)
Test navigation with text-based tree (no visuals). Measure success rate (≥70%), directness (≤1.5× optimal path), time. Identify problem areas, iterate.
Step 7: Implement and iterate (details)
Build high-fidelity designs and implement. Launch incrementally (pilot → rollout). Gather feedback from real users.
Step 8: Monitor findability metrics (details)
Track time to find, search success rate, navigation abandonment, bounce rate, user feedback. Refine taxonomy based on data.
设计或审计信息架构时，可采用以下结构化方法：
□ Step 1: Understand context and users
□ Step 2: Audit existing content (if any)
□ Step 3: Conduct user research (card sorting, interviews)
□ Step 4: Design taxonomy and navigation
□ Step 5: Create sitemap and wireframes
□ Step 6: Validate with tree testing
□ Step 7: Implement and iterate
□ Step 8: Monitor findability metrics步骤1：了解背景与用户（详情）
明确内容体量、用户目标、心智模型和成功指标（查找耗时、搜索查询、跳出率）。
步骤2：审计现有内容（详情）
盘点所有内容（URL、标题、元数据），识别重复内容、缺失内容和过时内容，评估当前性能（分析数据、热力图）。
步骤3：开展用户研究（详情）
邀请15-30位用户进行卡片分类（开放式、封闭式或混合式），分析聚类模式、分类标签和易混淆点，通过用户访谈理解其心智模型。
步骤4：设计分类体系与导航（详情）
构建层级结构（3-4层深度，每层5-9个选项），设计筛选维度，选择标签体系（任务导向、受众导向或字母排序），定义元数据架构。
步骤5：创建站点地图与线框图（详情）
通过可视化方式记录结构（站点地图示意图），创建包含导航、面包屑和筛选器的低保真线框图，收集利益相关者反馈。
步骤6：树状测试验证（详情）
使用基于文本的树状结构（无视觉元素）测试导航，衡量成功率（≥70%）、直接性（≤最优路径的1.5倍）和耗时，识别问题区域并迭代优化。
步骤7：落地与迭代（详情）
创建高保真设计并落地实现，分阶段发布（试点 → 全面推广），收集真实用户反馈。
步骤8：监控可查找性指标（详情）
追踪查找耗时、搜索成功率、导航放弃率、跳出率和用户反馈，基于数据优化分类体系。
Critical Guardrails
关键准则
1. Test with Real Users, Not Assumptions
1. 用真实用户测试，而非主观假设
Danger: Designing based on stakeholder opinions or personal preferences
Guardrail: Always validate with user research (card sorting, tree testing, usability testing). Minimum 15 participants for statistical significance.
Red flag: "I think users will understand 'Synergistic Solutions'..." — If you're guessing, you're wrong.
风险：基于利益相关者意见或个人偏好进行设计
准则：始终通过用户研究（卡片分类、树状测试、可用性测试）验证设计，最少15名参与者以保证统计显著性。
警示信号：「我认为用户能理解'协同解决方案'...」——如果只是猜测，结果往往是错误的。
2. Avoid Org Chart Navigation
2. 避免按组织架构设计导航
Danger: Structuring navigation by internal org structure (Sales, Marketing, Engineering)
Guardrail: Structure by user mental models and tasks, not company departments
Example: Bad: "About Us → Departments → Engineering → APIs". Good: "For Developers → APIs"
风险：按内部组织架构（销售、市场、工程）设计导航
准则：根据用户心智模型和任务设计结构，而非公司部门划分
示例：错误：「关于我们 → 部门 → 工程 → API」；正确：「开发者专区 → API」
3. Keep Navigation Shallow (3-4 Levels Max)
3. 保持导航层级较浅（最多3-4层）
Danger: Deep hierarchies (5+ levels) where users get lost
Guardrail: Aim for 3-4 levels deep, 5-9 items per level. If deeper needed, add search, filtering, or multiple entry points.
Rule of thumb: If users need >4 clicks from homepage to content, rethink structure.
风险：过深的层级（5+层）导致用户迷失
准则：目标为3-4层深度，每层5-9个选项。若需更深层级，需补充搜索、筛选或多个入口。
经验法则：若用户从首页到目标内容需点击4次以上，需重新设计结构。
4. Use Clear, Specific Labels (Not Jargon)
4. 使用清晰、具体的标签（避免行话）
Danger: Vague labels ("Resources", "Solutions") or internal jargon ("SKU Management")
Guardrail: Labels must be specific, action-oriented, and match user vocabulary. Test labels in card sorts and tree tests.
Test: Could a new user predict what's under this label? If not, clarify.
风险：模糊标签（如「资源」「解决方案」）或内部行话（如「SKU管理」）
准则：标签需具体、面向行动，并匹配用户词汇，在卡片分类和树状测试中验证标签的有效性。
测试方法：新用户能否预判该标签下的内容？若不能，需优化。
5. Ensure Single, Predictable Location
5. 确保内容位置唯一且可预判
Danger: Content lives in multiple places, or users can't predict location
Guardrail: Each content type should have ONE canonical location. If cross-category, use clear primary location + links from secondary.
Principle: "Principle of least astonishment" — content is where users expect it.
风险：内容存在多个位置，或用户无法预判其位置
准则：每种类型的内容应有一个标准位置。若需跨分类展示，需明确主位置并从次位置添加链接。
原则：「最小惊讶原则」——内容应在用户预期的位置。
6. Design for Scale
6. 为规模化增长设计
Danger: Structure works for 50 items but breaks at 500
Guardrail: Think ahead. If you have 50 products now but expect 500, design faceted navigation from start. Don't force retrofitting later.
Test: What happens if this category grows 10×? Will structure still work?
风险：结构在50条内容时可用，但在500条时失效
准则：提前规划。若当前有50个产品但预计增长到500个，从一开始就设计多面导航，避免后期返工。
测试方法：若该分类增长10倍，结构是否依然可用？
7. Provide Multiple Access Paths
7. 提供多种访问路径
Danger: Only one way to find content (e.g., only browse, no search)
Guardrail: Offer browse (navigation), search, filters, related links, breadcrumbs, tags. Different users have different strategies.
Principle: Some users are "searchers" (know what they want), others are "browsers" (exploring). Support both.
风险：仅提供一种查找内容的方式（如仅支持浏览，无搜索功能）
准则：提供浏览（导航）、搜索、筛选、相关链接、面包屑和标签等多种方式，不同用户有不同的查找策略。
原则：部分用户是「搜索者」（明确知道需求），部分是「浏览者」（探索式查找），需同时支持两种类型。
8. Validate Before Building
8. 先验证再开发
Danger: Building full site/app before testing structure
Guardrail: Use tree testing (text-based navigation) to validate structure before expensive design/dev work
ROI: 1 day of tree testing saves weeks of rework after launch.
风险：在测试结构前就完整开发网站/应用
准则：使用树状测试（基于文本的导航）在昂贵的设计和开发工作前验证结构
投资回报率：1天的树状测试可节省上线后数周的返工时间。
Quick Reference
快速参考
IA Methods Comparison
IA方法对比
| Method | When to Use | Participants | Deliverable |
|---|---|---|---|
| Open card sort | Exploratory, unknown categories | 15-30 users | Category labels, groupings |
| Closed card sort | Validation of existing categories | 15-30 users | Fit quality, confusion points |
| Tree testing | Validate navigation structure | 20-50 users | Success rate, directness, problem areas |
| Content audit | Understand existing content | 1-2 analysts | Inventory spreadsheet, gaps, duplicates |
| User interviews | Understand mental models | 5-10 users | Mental model diagrams, quotes |
| 方法 | 适用场景 | 参与人数 | 交付成果 |
|---|---|---|---|
| 开放式卡片分类 | 探索性研究，未知用户分类逻辑 | 15-30名用户 | 分类标签、内容分组 |
| 封闭式卡片分类 | 验证预设分类的合理性 | 15-30名用户 | 内容匹配度、易混淆点 |
| 树状测试 | 验证导航结构 | 20-50名用户 | 成功率、直接性、问题区域 |
| 内容审计 | 了解现有内容状况 | 1-2名分析师 | 内容盘点表、缺失/重复内容 |
| 用户访谈 | 理解用户心智模型 | 5-10名用户 | 心智模型图、用户语录 |
Navigation Depth Guidelines
导航深度指南
| Content Size | Recommended Structure | Example |
|---|---|---|
| <50 items | Flat (1-2 levels) | Blog, small product catalog |
| 50-500 items | Moderate (2-3 levels) | Documentation, medium e-commerce |
| 500-5000 items | Deep with facets (3-4 levels + filters) | Large e-commerce, knowledge base |
| 5000+ items | Hybrid (browse + search + facets) | Amazon, Wikipedia |
| 内容体量 | 推荐结构 | 示例 |
|---|---|---|
| <50条 | 扁平结构（1-2层） | 博客、小型产品目录 |
| 50-500条 | 中等深度（2-3层） | 文档、中型电商网站 |
| 500-5000条 | 深层+多面导航（3-4层+筛选器） | 大型电商网站、知识库 |
| 5000+条 | 混合结构（浏览+搜索+多面导航） | 亚马逊、维基百科 |
Labeling Systems
标签体系
| System | When to Use | Example |
|---|---|---|
| Task-based | Users have clear goals | "Book a Flight", "Track Order", "Pay Invoice" |
| Audience-based | Different user types | "For Students", "For Teachers", "For Parents" |
| Topic-based | Reference/learning content | "History", "Science", "Mathematics" |
| Format-based | Media libraries | "Videos", "PDFs", "Podcasts" |
| Alphabetical | No clear grouping, lookup-heavy | "A-Z Directory", "Glossary" |
| 体系类型 | 适用场景 | 示例 |
|---|---|---|
| 任务导向 | 用户有明确目标 | 「预订航班」「跟踪订单」「支付账单」 |
| 受众导向 | 存在不同用户群体 | 「学生专区」「教师专区」「家长专区」 |
| 主题导向 | 参考/学习类内容 | 「历史」「科学」「数学」 |
| 格式导向 | 媒体库 | 「视频」「PDF文档」「播客」 |
| 字母排序 | 无明确分组逻辑，以查找为主 | 「A-Z目录」「术语表」 |
Success Metrics
成功指标
| Metric | Target | Measurement |
|---|---|---|
| Tree test success rate | ≥70% | Users find correct destination |
| Directness | ≤1.5× optimal path | Clicks taken / optimal clicks |
| Time to find | <30 sec (simple), <2 min (complex) | Task completion time |
| Search success | ≥60% find without search | % completing task without search |
| Bounce rate | <40% | % leaving immediately from landing page |
| 指标 | 目标值 | 测量方式 |
|---|---|---|
| 树状测试成功率 | ≥70% | 用户找到正确目标的比例 |
| 直接性 | ≤最优路径的1.5倍 | 实际点击次数 / 最优点击次数 |
| 查找耗时 | 简单任务<30秒，复杂任务<2分钟 | 任务完成时间 |
| 搜索依赖度 | ≥60%的任务无需搜索即可完成 | 无需搜索完成任务的用户比例 |
| 跳出率 | <40% | 从着陆页立即离开的用户比例 |
Resources
资源
Navigation to Resources
资源导航
- Templates: Content audit template, card sorting template, sitemap template, tree testing script
- Methodology: Card sorting analysis, taxonomy design, navigation patterns, findability optimization
- Rubric: Evaluation criteria for IA quality (10 criteria)
- 模板：内容审计模板、卡片分类模板、站点地图模板、树状测试脚本
- 方法论：卡片分类分析、分类体系设计、导航模式、可查找性优化
- 评估标准：IA质量评估标准（10项指标）
Related Skills
相关技能
- data-schema-knowledge-modeling: For database schema and knowledge graphs
- mapping-visualization-scaffolds: For visualizing information structure
- discovery-interviews-surveys: For user research methods
- evaluation-rubrics: For creating IA evaluation criteria
- communication-storytelling: For explaining IA decisions to stakeholders
- data-schema-knowledge-modeling：用于数据库架构和知识图谱设计
- mapping-visualization-scaffolds：用于可视化信息结构
- discovery-interviews-surveys：用于用户研究方法
- evaluation-rubrics：用于创建IA评估标准
- communication-storytelling：用于向利益相关者解释IA决策
Examples in Context
场景示例
Example 1: E-commerce Navigation Redesign
示例1：电商导航重设计
Context: Bookstore with 10,000 books organized by publisher (internal logic)
Approach: Content audit → Open card sort (20 users: genre-based, not publisher) → Faceted navigation: Genre × Format × Price × Rating → Tree test (75% success) → Result: Time to find -40%, conversion +15%
背景：拥有10000本书籍的书店，按出版社（内部逻辑）组织内容
方案：内容审计 → 开放式卡片分类（20名用户：按流派而非出版社分组）→ 多面导航：流派 × 格式 × 价格 × 评分 → 树状测试（75%成功率）→ 结果：查找耗时减少40%，转化率提升15%
Example 2: SaaS Documentation IA
示例2：SaaS文档IA优化
Context: Developer docs, high abandonment after 2 pages
Approach: User interviews (mental model = tasks not features) → Taxonomy shift: feature-based to task-based ("Get Started", "Store Data") → Progressive disclosure (hub-and-spoke) → Tree test (68% → 82% success) → Result: Engagement +50%, support tickets -25%
背景：开发者文档，用户浏览2页后放弃的比例较高
方案：用户访谈（用户心智模型为任务而非功能）→ 分类体系从功能导向转为任务导向（「快速开始」「存储数据」）→ 渐进式披露（枢纽-分支结构）→ 树状测试（成功率从68%提升至82%）→ 结果：用户参与度提升50%，支持工单减少25%
Example 3: Internal Knowledge Base
示例3：内部知识库IA优化
Context: Company wiki with 2,000 articles, employees can't find policies
Approach: Content audit (40% outdated, 15% duplicates) → Closed card sort (25 employees) → Hybrid: browse (known needs) + search (unknown) + metadata schema → Search best bets → Result: Search success 45% → 72%, time to find 5min → 1.5min
背景：拥有2000篇文章的公司维基，员工无法找到政策类内容
方案：内容审计（40%内容过时，15%重复）→ 封闭式卡片分类（25名员工）→ 混合结构：浏览（已知需求）+ 搜索（未知需求）+ 元数据架构 → 搜索推荐 → 结果：搜索成功率从45%提升至72%，查找耗时从5分钟缩短至1.5分钟