SNA Chapter 1 Lecture 3

Summary

The lecture introduces common real-world network types, explains how networks can be built from different kinds of data, and previews the levels of analysis and topics covered in the course. Its central idea is that a network is an abstraction of a complex system: by representing entities as nodes and relationships as edges, researchers can study structures that may not initially look like networks.

Types of real-world networks

The lecturer groups networks into five broad categories:

• Social networks: People or accounts are nodes, connected by relationships such as friendship or following. Examples include Facebook friendship networks and Twitter follower–following networks.
• Biological networks: Examples include protein–protein interactions, metabolic networks linking metabolites and enzymes, neural connections, and food networks.
• Information networks: Examples include the World Wide Web and citation networks.
• Technological networks: Examples include power grids, airline and railway systems, and communication networks involving computers, routers, or satellites.
• Language networks: Words or keywords can be connected when they frequently occur together.

Networks can also be constructed for many other settings, including terrorism-related activity and patient contact during an epidemic.

How different networks are represented

A network’s nodes and edges depend on the system being studied. Examples include:

• Citation networks: Papers are nodes, and citations are directed edges. Incoming edges represent citations received, while outgoing edges represent references made. A paper with many incoming citations may be influential; many outgoing edges may indicate a work that draws on a large body of literature, such as a review.
• Co-authorship networks: Authors are nodes, connected when they have worked together.
• Word co-occurrence networks: Words are nodes, connected when they appear near one another—for example, within a chosen window. These networks can help identify linguistic relationships, including synonyms or antonyms.
• Keyword co-occurrence networks: Keywords from papers are connected when they repeatedly appear together, revealing related research topics.
• Semantic networks or knowledge graphs: Entities are nodes, and relationships such as “is a” or “has” are labeled edges. For example, “cat” can be connected to “mammal” by an “is a” relationship.
• Patient networks: Patients are nodes, connected when they interact or come into close contact. Such a network can help trace possible paths of disease transmission and identify people who may need protection.

Three levels of network analysis

The lecturer describes three levels of granularity:

1. Microscopic analysis — individual nodes and local relationships

   • Studies node and edge properties, such as degree or centrality.
   • Examines interactions between pairs of nodes, called dyadic interactions.
   • Examines interactions among three nodes, called triadic interactions.
   • Can focus on an ego network: one focal node (the ego), its immediate neighbors (alters), and the relationships among those neighbors. Closely connected groups within an ego network may be called circles.

2. Mesoscopic analysis — structures within the network

   • Focuses on regions or recurring substructures rather than individual nodes or the entire network.
   • Communities or clusters are groups whose members are densely connected to one another.
   • Network motifs are recurring small patterns, such as stars or chains. Different motif patterns can help distinguish networks and are especially useful in biological network analysis.

3. Macroscopic analysis — the network as a whole

   • Looks at overall properties, including node count, degree distribution, edge density, and diameter.
   • Diameter is the longest among the shortest paths between pairs of nodes.
   • Edge density compares the number of edges present with the number that could potentially exist.

Broad questions and properties in network science

A key question is whether different kinds of networks share structural properties. Similarities between, for example, a social network and a protein interaction network do not mean they are the same system; they may instead point to general principles of network structure, evolution, or dynamics.

The lecture previews several properties and phenomena:

• Small-world structure: Nodes may be connected through relatively short paths. The lecturer refers to the “six degrees of separation” idea.
• Scale-free structure
• Clustering and community structure
• Robustness to attacks
• Cascading failures: A failure at one point in a network, such as a power grid, may propagate and disrupt other parts of the system.

Applications and course topics

The lecture highlights both analytical applications and the planned course sequence:

• Node classification: Assigning categories to nodes based on their properties.
• Link prediction and recommendation: Predicting future or missing relationships, with applications such as friendship, product, and advertising recommendations.
• Growth and virality: Studying how networks or messages spread and expand.
• Misinformation and epidemic spread: Analyzing cascades through online or offline networks.
• Anomaly detection: Identifying unusual nodes, edges, or subgraphs, including potential fraud or outlier activity.
• Network formation models: Exploring how networks form and evolve, including random-growth and preferential-attachment models.
• Link analysis and community detection: Characterizing relationships and finding densely connected groups.
• Graph or network representation learning: Mapping a network into a vector space so that machine-learning operations can be applied to its elements.
• Fraud and collusion detection: Examining suspicious users, coordinated activity, and black-market-driven behavior in online networks.
• Recommender systems: Including friendship recommendations.

Suggested background

The lecturer recommends familiarity with Python, introductory probability and statistics, basic linear algebra (especially matrix operations), and basic algorithm design. Introductory machine learning and deep learning would also be useful for the representation-learning material.

Speakers and sources featured

• Speaker: One lecturer or instructor; not identified by name in the subtitles.
• Other speakers: None identified.