
课程介绍

图是一种强大的数据结构,可以用于建模许多真实世界的场景,图能够对样本之间的关系信息进行建模。但是真实图的数据量庞大,动辄上亿节点、而且内部拓扑结构复杂,很难将传统的图分析方法如最短路径、DFS、BFS、PageRank 等算法应用到这些任务上。因此有研究者提出将机器学习方法和图数据结合起来,即图机器学习,这逐渐成为近年来机器学习中的一股热潮,特别是图神经网络(GNN)。

CS224W 是顶级院校斯坦福出品的图机器学习方向专业课程,对于graph方向的数据挖掘和机器学习(神经网络)有全面的知识覆盖和很高的权威度。如果大家想学习非结构化的图数据上的各类算法,本课程是最适合的课程之一。

课程主讲人 Jure Leskovec 是斯坦福大学计算机科学副教授,也是图表示学习方法 node2vec 和 GraphSAGE 的作者之一。
他主要的研究兴趣是社会信息网络的挖掘和建模等,特别是针对大规模数据、网络和媒体数据。多次在 Nature、NeurIPS、KDD、ICML 等期刊和学术会议上发表论文,并两次获得 KDD 时间检验奖。
课程主题
本课程着重于分析海量图形所面临的计算、算法和建模挑战,通过研究底层图结构及其特征,向学生介绍机器学习技术、数据挖掘工具。课程涉及的主题包括:
- Machine Learning for Graphs(基于图的机器学习)
- Traditional Methods for ML on Graphs(图数据上的传统方法)
- Node Embeddings(节点嵌入)
- Link Analysis: PageRank(PageRank)
- Label Propagation for Node Classification(用于节点分类的标签传播)
- Graph Neural Networks(图神经网络)
- Knowledge Graph Embeddings(知识图谱嵌入)
- Reasoning over Knowledge Graphs(基于知识图的推理)
- Frequent Subgraph Mining with GNNs(使用GNN进行频繁子图挖掘)
- Community Structure in Networks(网络中的社区结构)
- Traditional Generative Models for Graphs(图数据的传统生成模型)
- Deep Generative Models for Graphs(图数据的深度生成模型)
- Advanced Topics on GNNs(GNN 进阶专题)
- Scaling Up GNNs(大规模GNN)
- Guest Lecture: GNNs for Computational Biology(GNNs在计算生物学的应用)
- Guest Lecture: Industrial Applications of GNNs(GNNs的工业应用)
- GNNs for Science(用于科学的 GNN)
课程资料 | 下载
![]() |
扫描上方图片二维码,关注公众号并回复关键字 🎯『CS224W』,就可以获取整理完整的资料合辑啦!当然也可以点击 🎯 这里 查看更多课程的资料获取方式!


ShowMeAI 对课程资料进行了梳理,整理成这份完备且清晰的资料包:
- 📚 课件(PDF)。Lecture 1~20所有章节,图文结合的呈现,对于理解很有帮助。- 📚 代码及作业参考解答(.ipynb)。Colab 0~4代码,Homework 0~3作业答案。- 📚 拓展阅读 & 知识图谱资源大全。课程推荐的相关学习资源、清单。
课程视频 | B站
ShowMeAI 将视频上传至B站,并增加了中英双语字幕,以提供更加友好的学习体验。点击页面视频,可以进行预览。推荐前往 👆 B站 观看完整课程视频哦!
本门课程,ShowMeAI 将部分章节进行了切分,按照主题形成更短小的视频片段,便于按照标题进行更快速的检索。切分后的视频清单列写在这里
- 注意:章节名术语较多,未作中文翻译
序号 | 视频章节 | 视频清单 |
---|---|---|
L1 | L1.1 | Why Graphs |
L2 | L1.2 | Applications of Graph ML |
L3 | L1.3 | Choice of Graph Representation |
L4 | L2.1 | Traditional Feature-based Methods- Node |
L5 | L2.2 | Traditional Feature-based Methods- Link |
L6 | L2.3 | Traditional Feature-based Methods_ Graph |
L7 | L3.1 | Node Embeddings |
L8 | L3.2 | Random Walk Approaches for Node Embeddings |
L9 | L3.3 | Embedding Entire Graphs |
L10 | L4.1 | PageRank |
L11 | L4.2 | PageRankHow to Solve |
L12 | L4.3 | Random Walk with Restarts |
L13 | L4.4 | Matrix Factorizing and Node Embeddings |
L14 | L5.1 | Message passing and Node Classification |
L15 | L5.2 | Relational and Iterative Classification |
L16 | L5.3 | Collective Classification |
L17 | L6.1 | Graph Neural Networks Introduction |
L18 | L6.2 | Basics of Deep Learning |
L19 | L6.3 | Deep Learning for Graphs |
L20 | L7.1 | A General Perspective on GNN |
L21 | L7.2 | A Single Layer of a GNN |
L22 | L7.3 | Stacking layers of a GNN |
L23 | L8.1 | Graph Augmentation for GNNs |
L24 | L8.2 | Training Graph Neural Networks |
L25 | L8.3 | Setting up GNN Prediction Tasks |
L26 | L9.1 | How Expressive are Graph Neural Networks |
L27 | L9.2 | Designing the Most Powerful GNNs |
L28 | L10.1 | Heterogeneous & Knowledge Graph Embedding |
L29 | L10.2 | Knowledge Graphs KG Completion |
L30 | L10.3 | Knowledge Graph Completion |
L31 | L11.1 | Reasoning in Knowledge Graphs |
L32 | L11.2 | Answering Predictive Queries |
L33 | L11.3 | Query2box Reasoning over KGs |
L34 | L12.1 | Fast Neural Subgraph Matching & Counting |
L35 | L12.2 | Neural Subgraph Matching |
L36 | L12.3 | Finding Frequent Subgraphs |
L37 | L13.1 | Community Detection in Networks |
L38 | L13.2 | Network Communities |
L39 | L13.3 | Louvain Algorithm |
L40 | L13.4 | Detecting Overlapping Communities |
L41 | L14.1 | Generative Models for Graphs |
L42 | L14.2 | Erdos Renyi Random Graphs |
L43 | L14.3 | The Small World Model |
L44 | L14.4 | Kronecker Graph Model |
L45 | L15.1 | Deep Generative Models for Graphs |
L46 | L15.2 | Graph RNN Generating Realistic Graphs |
L47 | L15.3 | Scaling Up & Evaluating Graph Gen |
L48 | L15.4 | Application of Deep Graph Generative |
L49 | L16.1 | Limitations of Graph Neural Networks |
L50 | L16.2 | Position aware Graph Neural Networks |
L51 | L16.3 | Identity-Aware Graph Neural Networks |
L52 | L16.4 | Robustness of Graph Neural Networks |
L53 | L17.1 | Scaling Up Graph Neural Networks to Large Graphs |
L54 | L17.2 | GraphSAGE Neighbor Sampling |
L55 | L17.3 | Cluster GCN Scaling up GNNs |
L56 | L17.4 | Scaling up by Simplifying GNNs |
L57 | L18 | GNNs in Computational Biology |
L58 | L19.1 | Pre Training Graph Neural Networks |
L59 | L19.2 | Hyperbolic Graph Embeddings |
L60 | L19.3 | Design Space of Graph Neural Networks |
根据视频内容整理的这份『CS224W 课程结构图解』,展示了内容要点及其逻辑关系,超级直观!相信对构建 Whole Picture 特别有帮助~

学习建议
- 这门课程是独立的。
- 单个Topic容易理解,但课程涉及到了很多Topic,这就导致了课程难度的增加。
- 学生应具备以下背景:
- 具备基本的计算机科学原理知识,足以编写合理的非琐碎计算机程序
- 熟悉基本的概率理论和线性代数
- 熟悉机器学习、算法与图论的基本知识
- 课程前几周的复习课将概述预期的背景
更多技术与课程清单 | 点击查看详细课程
