Course team

Charismatic lecturer

Helpfull assistant

Ilia Karpov

• 24 Scopus papers
• 30 Projects in NLP, SNA
• CTO in 2 companies
• Cofounder in 4 companies

Форматы и технологии

Usefull links

• NotebookLM
• Course program
• Telegram channel
• Grading table

Historical Prospective

Complex Networks

• not regular, but not random
• non-trivial topology
• universal properties
• everywhere
• complex systems

Complex Networks

• power law node degree distribution: ”scale-free” networks
• small diameter and average path length: ”small world” networks
• hight clustering coefficient: transitivity

Power law

Power law

Power Law

Triads

• ”The Strength of Weak Ties”, Mark Grannoveter, 1973
• ”Spread of Information through a Population with Socio-Structural Bias. Assumption of Transitivity”, Anatol Rapoport, 1953

Triads

If A and B and B and C are strongly linked, the the tie between B and C is always present

Grannoveter, 1973

• strength of a tie
• high transitivity
• high clustering coefficient

High Clustering Coefficient

• ”The small-world problem”. Stanley Milgram, 1967, 10871📝 [pdf]
• ”An experimental study of the small world problem”, Jeffrey Travers, Stanley Milgram, 1969, 3405 📝, [pdf]

Stanley Milgram's experiment

Stanley Milgram's experiment

Stanley Milgram's experiment

• Reached the target N = 64(29%)
• Average chain length (L) = 5.2
• Channels:
  • hometown (L) = 6.1
  • business contacts (L) = 4.6
  • from Boston (L) = 4.4
  • from Nebraska (L) = 5.7

Stanley Milgram's experiment

• Email graph:
  • D. Watts (2001), 48,000 senders, (L) ≈ 6
• MSN Messenger graph:
  • J. Lescovec et al (2007), 240mln users, (L) ≈ 6.6
• Facebook graph:
  • L. Backstrom et al (2012), 721 mln users, (L) ≈ 4.74

Spring of Neural Networks

Shallow embeddings

• DeepWalk: online learning of social representations, KDD 14,  Cited by 10349, [pdf]

Complex Embeddings

• Semi-Supervised Classification with Graph Convolutional Networks,  ICLR17, Cited by 30292, [pdf]

GraphRag

• From Local to Global: A Graph RAG Approach to Query-Focused Summarization, [pdf]

Содержание курса (1/3)

1. Введение в Network Science — задачи, источники данных; практика: Gephi.
2. Базовые определения I (метрики целого графа) — пути, диаметр, кластеризация.
3. Инструменты анализа — форматы графов и тулчин.
4. Базовые определения II (метрики вершин) — betweenness, closeness, PageRank, degree, cores, motifs, HITS, двудольные графы. Практика: NetworkX/igraph.
5. Community Detection — модульность, спектральные методы, Лувен/Лейден; практика: Python.

Содержание курса (2/3)

6. Block Modeling — типы, алгоритмы, матрица изображений и интерпретация.
7. Сетевые модели — Эрдёш–Реньи, preferential attachment, small-world; практика: генерация графов.
8. Распространение/эпидемиология — SI/SIS/SIR, слухи; практика: ndlib.
9. Графовые эмбеддинги — unsupervised, factorization, random walk; практика: link prediction.
10. GNN I — message passing, DGL; практика: классификация узлов.

Содержание курса (3/3)

11. GNN II — link prediction, классификация графов, SOTA особенности.
12. Текстовые графы и KGs — анализ сообщений, JODIE; практика: Entity Linking в DGL.
13. Поиск на графах : FAISS, Gensim, Elasticsearch, RAG.
14. Темпоральные графы — модели, структуры данных, динамические процессы.
15. Гостевая лекция — применимость SNA (рекомендательные системы и др.).

1. Introduction to Network Science, Tools for Graph Analysis and Visualization

Theory:

• Tasks and applications of network analysis
• Key ideas in network analysis from its origins to recent advancements
• Overview of the most authoritative journals and conferences
• Practical examples of network analysis applications
  Brief review of available data sources for graph construction

Practice

• Gephi

2. Basic Definitions in Network Analysis: Whole Graph Characteristics, Node-Level Characteristics

Theory: 

• Concepts of graph, vertices, edges, incidence matrix, connectivity, vertex closeness, diameter, graph paths, and average path length.
• betweenness, closeness, PageRank, degree, cores, graph motifs, bipartite networks, HITS

Practice

• NetworkX
• IGraph
• Familiarization with graph processing packages (Python)
• Practical application of node-level metrics
• Calculating node characteristics in graphs
  
   

3. Community Detection

Theory: 

• Cohesive subgroups, graph cliques, network communities, graph partitioning, modularity, boundary betweenness, spectral partitioning, modularity maximization, heuristic methods, label propagation, rapid community evolution, walktrap algorithm.
• Characteristics of empirical models and generative models.
• Random graph model, phase transition, growing random graphs, preferential attachment model, small-world model, Erdős–Rényi model, preferential embedding

Practice

• Data collection and preparation for analysis.
• Community detection and clustering in Python.
• Generating random graphs

4. Information Propagation and Epidemiological Models on Graphs

Theory: 

• Epidemic models on networks.
• SI, SIS, and SIR models, rumor spreading, propagation trees

Practice

• Epidemic modeling using ndlib

5. Graph Embeddings

Theory: 

• Unsupervised graph embeddings, matrix factorization, random walk methods

Practice

• Link prediction using unsupervised embeddings

6. Graph Neural Networks
 

Theory: 

• Message passing model
• Overview of libraries such as Deep Graph Library (DGL) and PyTorch Geometric (PyG)
• GIANT model, Weisfeiler-Lehman algorithm, and analysis of large graphs

Practice

• Node classification using DGL
• Whole graph classification
• Features of SOTA (State-of-the-Art) classification algorithms

7. Graphs in Recommender Systems, Temporal Graphs
 

Theory: 

• Problem formulation for recommender systems
• Models of temporal networks
• Data structures and encoding
• Measurements and properties of temporal networks
• Dynamic processes in temporal networks, algorithms, and analysis methods

Practice

• Implementation of LightGCN

8. Text Graphs and Knowledge Graphs, Graphs in Search Systems

Theory: 

• Construction of knowledge graphs
• Graph-based search
•  HNSW
•  RAG
•  GraphRag

Practice

• Entity linking using DGL
• Familiarization with information retrieval tools such 
• Fais
• Gensim
• Elasticsearch

Course grading

if \( Grade \) >= 6, student can skip exam

else:

where N - is the number of quizes;

W - is the number of contests.

Retaking the exam

  The procedure for the first retake and the thematic composition of questions are similar to the exam. The score is given at the end of the meeting.

   The procedure for the second retake and the thematic composition of questions are similar to the exam. The retake is carried out in the presence of at least three members of the commission, including its chairman. The score is given at the end of the meeting. The score for the second retake is set by agreement of the members of the commission. In case of divergence of assessments, the decision is made by a majority of votes, in case of equality of votes, the assessment of the chairman of the commission is decisive. Blocking elements are not provided.

Course logistics

• Lectures and Practice are held each Thursday 18:10, D-501;
• Discussions and Announcements in telegram chat;

• All deadlines are 23:59 pm anywhere on earth

• Consulting hours booking page

Система оценивания — формулы

Итог: G = 0.5 · G_hw + 0.5 · G_pr

Экзамен / автомат

• Если G ≥ 6.0 — возможен автомат, равный накопленной оценке.
• Если G < 6.0 — устный экзамен (3 вопроса).

Оценка за экзамен: G_ex = 0.6 · questions(mark) + 0.4 · G

Точные формулировки вопросов см. в списке экзаменационных тем.

SNA Coins 2025

Как это работает?

• За активность и результаты вы получаете SNA Coins.
• Баллы можно тратить индивидуально или в складчину.
• Учёт баллов: публичный Google Sheet.

Как заработать?

Анти-фарм правила

• Вопросы с баллами: 2 за семинар.
• Обзор статьи: ≤2 на студента за семестр.
• Дубли активности не считаются.

Прозрачность

• Открытый рейтинг: /leaderboard.
• История начислений в Google Sheet.

На что тратить?