LightRAG: 简单快速的检索增强生成
✨步子哥 (steper) •
2025年11月29日 09:38
<!DOCTYPE html>
<html lang="zh">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>LightRAG: 简单快速的检索增强生成</title>
<link href="https://fonts.googleapis.com/icon?family=Material+Icons" rel="stylesheet">
<link href="https://fonts.googleapis.com/css2?family=Noto+Sans+SC:wght@400;500;700&display=swap" rel="stylesheet">
<style>
* {
margin: 0;
padding: 0;
box-sizing: border-box;
}
body {
font-family: 'Noto Sans SC', sans-serif;
background-color: #f5f7fa;
color: #333;
line-height: 1.6;
}
.poster-container {
width: 720px;
min-height: 960px;
margin: 0 auto;
background: linear-gradient(135deg, #e6f1ff 0%, #f0f4f8 100%);
padding: 40px;
position: relative;
overflow: hidden;
}
.bg-accent {
position: absolute;
width: 400px;
height: 400px;
border-radius: 50%;
background: radial-gradient(circle, rgba(64, 169, 255, 0.1) 0%, rgba(64, 169, 255, 0) 70%);
z-index: 0;
}
.bg-accent-1 {
top: -100px;
right: -100px;
}
.bg-accent-2 {
bottom: -150px;
left: -150px;
}
.header {
text-align: center;
margin-bottom: 30px;
position: relative;
z-index: 1;
}
.title {
font-size: 50px;
font-weight: 700;
color: #1a73e8;
margin-bottom: 15px;
line-height: 1.2;
}
.subtitle {
font-size: 24px;
color: #5f6368;
max-width: 90%;
margin: 0 auto;
}
.section {
background-color: white;
border-radius: 16px;
padding: 25px;
margin-bottom: 25px;
box-shadow: 0 4px 12px rgba(0, 0, 0, 0.05);
position: relative;
z-index: 1;
}
.section-title {
display: flex;
align-items: center;
font-size: 28px;
font-weight: 700;
color: #1a73e8;
margin-bottom: 15px;
}
.section-title .material-icons {
margin-right: 10px;
font-size: 32px;
}
.dilemma-container {
display: flex;
justify-content: space-between;
margin-top: 20px;
}
.dilemma-card {
width: 48%;
padding: 20px;
border-radius: 12px;
background-color: #f8f9fa;
}
.dilemma-card h3 {
font-size: 20px;
margin-bottom: 10px;
color: #1a73e8;
}
.dilemma-card p {
font-size: 16px;
color: #5f6368;
}
.dilemma-arrow {
display: flex;
align-items: center;
justify-content: center;
font-size: 24px;
color: #ea4335;
}
.architecture {
display: flex;
flex-direction: column;
gap: 20px;
}
.arch-step {
display: flex;
align-items: flex-start;
gap: 15px;
}
.step-number {
display: flex;
align-items: center;
justify-content: center;
width: 36px;
height: 36px;
border-radius: 50%;
background-color: #1a73e8;
color: white;
font-weight: 700;
flex-shrink: 0;
}
.step-content {
flex-grow: 1;
}
.step-content h3 {
font-size: 18px;
margin-bottom: 5px;
color: #1a73e8;
}
.step-content p {
font-size: 16px;
color: #5f6368;
}
.comparison-table {
width: 100%;
border-collapse: collapse;
margin-top: 15px;
}
.comparison-table th, .comparison-table td {
padding: 12px 15px;
text-align: left;
border-bottom: 1px solid #e0e0e0;
}
.comparison-table th {
background-color: #f1f3f4;
font-weight: 500;
color: #1a73e8;
}
.highlight {
background-color: #e8f0fe;
padding: 2px 5px;
border-radius: 4px;
font-weight: 500;
}
.advantage {
display: flex;
align-items: center;
margin-bottom: 10px;
}
.advantage .material-icons {
color: #34a853;
margin-right: 10px;
}
.use-cases {
display: flex;
flex-wrap: wrap;
gap: 15px;
margin-top: 15px;
}
.use-case {
flex: 1 0 30%;
background-color: #f1f3f4;
padding: 15px;
border-radius: 8px;
display: flex;
align-items: center;
}
.use-case .material-icons {
color: #1a73e8;
margin-right: 10px;
}
.use-case p {
font-size: 16px;
color: #5f6368;
}
.image-container {
text-align: center;
margin: 20px 0;
}
.image-container img {
max-width: 100%;
border-radius: 8px;
box-shadow: 0 4px 8px rgba(0, 0, 0, 0.1);
}
.dual-retrieval {
display: flex;
justify-content: space-between;
margin-top: 20px;
gap: 20px;
}
.retrieval-level {
flex: 1;
padding: 15px;
border-radius: 8px;
background-color: #f8f9fa;
}
.retrieval-level h3 {
font-size: 18px;
margin-bottom: 10px;
color: #1a73e8;
display: flex;
align-items: center;
}
.retrieval-level h3 .material-icons {
margin-right: 8px;
}
.retrieval-level p {
font-size: 16px;
color: #5f6368;
}
</style>
</head>
<body>
<div class="poster-container">
<div class="bg-accent bg-accent-1"></div>
<div class="bg-accent bg-accent-2"></div>
<header class="header">
<h1 class="title">LightRAG: 简单快速的检索增强生成</h1>
<p class="subtitle">跳出"二选一"困境，同时具备图谱的深度和向量检索的速度</p>
</header>
<section class="section">
<h2 class="section-title">
<i class="material-icons">help_outline</i>
RAG系统的两难问题
</h2>
<div class="dilemma-container">
<div class="dilemma-card">
<h3>传统RAG</h3>
<p>使用简单向量检索，成本低、速度快，但答案常常碎片化，无法理解复杂关联</p>
</div>
<div class="dilemma-arrow">
<i class="material-icons">compare_arrows</i>
</div>
<div class="dilemma-card">
<h3>GraphRAG</h3>
<p>能理解数据里复杂的关联，但构建成本高昂，更新困难，检索速度慢</p>
</div>
</div>
</section>
<section class="section">
<h2 class="section-title">
<i class="material-icons">lightbulb</i>
LightRAG的核心思想
</h2>
<p>LightRAG提出了一种轻量级的思路，核心不是去建造一个庞大臃肿的知识图谱，而是用一种更巧妙、更经济的方式，让系统同时具备图谱的深度和向量检索的速度。</p>
<div class="image-container">
<img src="https://sfile.chatglm.cn/moeSlide/image/fd/fd8087af.jpg" alt="LightRAG架构图">
</div>
<div class="dual-retrieval">
<div class="retrieval-level">
<h3><i class="material-icons">filter_1</i> 低层检索</h3>
<p>聚焦实体属性与直接关系，精确检索特定节点或边的信息</p>
</div>
<div class="retrieval-level">
<h3><i class="material-icons">filter_2</i> 高层检索</h3>
<p>聚合跨实体的全局主题，提供对更高层次概念和摘要的洞察</p>
</div>
</div>
</section>
<section class="section">
<h2 class="section-title">
<i class="material-icons">architecture</i>
LightRAG的架构设计
</h2>
<div class="architecture">
<div class="arch-step">
<div class="step-number">1</div>
<div class="step-content">
<h3>实体和关系提取</h3>
<p>使用LLM从文本块中提取实体(节点)和关系(边)</p>
</div>
</div>
<div class="arch-step">
<div class="step-number">2</div>
<div class="step-content">
<h3>LLM生成键值对</h3>
<p>为每个实体和关系生成文本键值对，键是检索索引词，值是总结性文本</p>
</div>
</div>
<div class="arch-step">
<div class="step-number">3</div>
<div class="step-content">
<h3>图谱去重</h3>
<p>合并不同文本块中的相同实体和关系，减小图规模</p>
</div>
</div>
<div class="arch-step">
<div class="step-number">4</div>
<div class="step-content">
<h3>增量更新</h3>
<p>新文档独立处理后与主图谱合并，无需重建整个索引</p>
</div>
</div>
</div>
</section>
<section class="section">
<h2 class="section-title">
<i class="material-icons">compare</i>
LightRAG vs GraphRAG
</h2>
<table class="comparison-table">
<tr>
<th>比较维度</th>
<th>LightRAG</th>
<th>GraphRAG</th>
</tr>
<tr>
<td>检索成本</td>
<td><span class="highlight">极低 (< 100 token)</span></td>
<td>高昂 (610,000 token)</td>
</tr>
<tr>
<td>更新方式</td>
<td><span class="highlight">增量更新</span></td>
<td>全量重建</td>
</tr>
<tr>
<td>检索速度</td>
<td><span class="highlight">快速</span></td>
<td>较慢</td>
</tr>
<tr>
<td>答案多样性</td>
<td><span class="highlight">高 (76.4% vs 23.6%)</span></td>
<td>低</td>
</tr>
<tr>
<td>答案全面性</td>
<td><span class="highlight">高 (83.6% vs 16.4%)</span></td>
<td>低</td>
</tr>
</table>
<div class="image-container">
<img src="https://sfile.chatglm.cn/moeSlide/image/1e/1edc2493.jpg" alt="LightRAG流程图">
</div>
</section>
<section class="section">
<h2 class="section-title">
<i class="material-icons">stars</i>
LightRAG的优势
</h2>
<div class="advantage">
<i class="material-icons">check_circle</i>
<p>检索成本降低99%，同时保持高质量的答案生成</p>
</div>
<div class="advantage">
<i class="material-icons">check_circle</i>
<p>支持增量更新，适合动态数据环境</p>
</div>
<div class="advantage">
<i class="material-icons">check_circle</i>
<p>双层检索机制，兼顾细节与全局信息</p>
</div>
<div class="advantage">
<i class="material-icons">check_circle</i>
<p>在多个评估维度上全面优于现有方法</p>
</div>
</section>
<section class="section">
<h2 class="section-title">
<i class="material-icons">category</i>
适用场景
</h2>
<div class="use-cases">
<div class="use-case">
<i class="material-icons">update</i>
<p>动态数据环境</p>
</div>
<div class="use-case">
<i class="material-icons">speed</i>
<p>高效检索需求</p>
</div>
<div class="use-case">
<i class="material-icons">hub</i>
<p>跨领域综合查询</p>
</div>
<div class="use-case">
<i class="material-icons">question_answer</i>
<p>智能客服系统</p>
</div>
<div class="use-case">
<i class="material-icons">analytics</i>
<p>企业知识管理</p>
</div>
<div class="use-case">
<i class="material-icons">school</i>
<p>学术研究辅助</p>
</div>
</div>
</section>
</div>
</body>
</html>
登录后可参与表态
讨论回复
0 条回复还没有人回复，快来发表你的看法吧！