Social Network Analysis | Chapter 1 | Networks and Society

Summary

The lecture introduces social network analysis (SNA) as the study of relationships and patterns in social systems using networks, or graphs. Networks provide a simplified way to represent complex systems and make their structure and behavior easier to analyze.

What a Network Represents

A network consists of nodes (also called vertices or entities) and edges (also called links or connections). Formally, it can be represented as (G = (V, E)), where (V) is the set of nodes and (E) is the set of edges.

Nodes can represent people, computers, web pages, researchers, products, posts, or other entities. Edges represent relationships or interactions between them. Networks can model online activity—such as messages, follows, and likes—as well as offline relationships, such as people meeting or communicating.

Networks are useful abstractions of complex systems, including biological systems, the World Wide Web, and social groups. Large real-world networks are often dynamic: nodes and links may appear or disappear, and their structure can be too complex to understand through visual inspection alone. Studying their structural and functional properties is the focus of network science and complex-network research.

Common Network Types and Modeling Choices

When modeling a system as a network, an analyst must decide what its nodes and connections represent, and which properties to preserve.

• Undirected networks: Connections have no direction and are symmetric, as in a mutual friendship relationship.
• Directed networks: Connections point from one node to another and may not be reciprocal, as with subscriptions or follows.
• Weighted networks: Edges carry values, such as interaction frequency, message volume, or traffic.
• Loops and parallel edges: A loop connects a node to itself; parallel edges represent multiple connections between the same pair of nodes.
• Bipartite networks: Nodes belong to two groups, with edges only between groups. A user–product purchase network is one example. Networks can also have three or more partitions.
• Signed networks: Edges have positive, negative, or sometimes neutral signs, representing relationships such as trust or hostility. These signs can help analyze social stability and balance.
• Homogeneous and heterogeneous networks: A homogeneous network has one type of node and one type of relationship. A heterogeneous network can contain different node types, relationship types, or both—for example, users and posts connected through follows, replies, retweets, or likes.
• Attribute networks: Nodes or edges carry additional information, such as a user’s age or profession, or the content of a message.
• Multilayer networks: Different kinds of relationships are represented in separate layers, such as one layer for follows and another for retweets.
• Temporal networks: The network is tracked as it changes over time.
• Egocentric networks: A local view centered on one node, called the ego, and its directly connected neighbors, called alters. The ego network includes the connections among those neighbors as they appear in the original network.
• Hypergraphs: A hyperedge can connect more than two nodes. For example, it can represent a paper jointly written by several researchers.

Representing Networks in Data Structures

An adjacency matrix records which pairs of nodes are connected. It is symmetric for an undirected network and generally asymmetric for a directed one.

For large, sparse networks, matrices may be inefficient because most entries are zero. An adjacency list, which stores each node alongside its neighbors, is often more practical for such networks.

Scales of Network Analysis

The lecture describes three levels of analysis:

• Microscopic: Examines individual nodes, edges, and small interactions. A dyadic interaction involves two nodes; a triadic interaction involves three. Egocentric networks also provide a local, microscopic view.
• Mesoscopic: Examines structures between individual interactions and the network as a whole, such as communities, clusters, and recurring subgraphs.
• Macroscopic: Examines the network as a whole. Examples include average path length, diameter, and edge density. The diameter is the longest among the shortest paths between node pairs; density compares the number of existing edges with the number of possible edges.

A network motif is a small subgraph that occurs repeatedly. Motif patterns—such as chains, stars, or cliques—can reveal features of a network. The lecture notes their particular relevance in areas such as biology and collaboration analysis.

Applications of Social Network Analysis

• Healthcare and epidemics: Contact networks can help identify potentially exposed people, anticipate disease spread, and inform vaccination, isolation, or lockdown strategies.
• Recommendations: Social connections can help explain friend suggestions on social platforms. Network structure also supports product recommendations in e-commerce.
• Information diffusion and marketing: Analysis can trace how information, propaganda, misinformation, or other content spreads. Network analysis can also help select influential people to promote a product or message within a limited budget.
• Web search: The World Wide Web can be modeled as a graph of pages connected by hyperlinks. The lecture uses PageRank as an example of a graph-based method for ranking web pages.
• Cybercrime and harmful content: Network models can help study fraud, fake news, cyberbullying, trolling, and offensive content, and identify ways to limit their spread.
• Security and terrorism research: The lecture describes the use of interaction networks to identify important participants in terrorist networks.
• Science and research: Citation networks connect articles through citations, while collaboration networks connect researchers who have worked together. Both can help track how research develops and suggest literature or potential collaborators.
• Infrastructure: Networks such as power grids can be studied for resilience, including how to prevent a local failure from cascading into wider disruption.

Patterns Commonly Studied in Real-World Networks

The lecture highlights several recurring properties and research questions:

• Many real-world networks have short paths between nodes, associated with the “six degrees” or small-world idea.
• Many are described as having power-law degree distributions and scale-invariant properties.
• Similar nodes often cluster together, forming communities.
• Researchers study how robust a network is to random failures and how targeted removals of nodes or edges may disrupt its structure.
• Cascades occur when an event or piece of information propagates through connected nodes—for example, a chain of retweets or a failure spreading through a power grid.

Why the Field Matters and Common Analysis Tasks

The growth of social media has made large amounts of interaction data available. This data can support research into how people behave and interact online, while also creating the challenge of analyzing networks too large to inspect directly.

Common tasks include:

• Node classification: Determine a node’s category, such as distinguishing scam accounts from legitimate accounts.
• Relationship prediction: Predict likely connections, supporting friend suggestions and other recommendations.
• Virality prediction: Estimate whether a product, message, or post may spread widely.
• Importance or authority detection: Identify influential nodes in a network or community.
• Sampling: Analyze a manageable portion of a very large network when the entire graph cannot be accessed or processed.
• Information-spread analysis: Understand and potentially limit the spread of misinformation, hate speech, and other harmful content.
• Anomaly detection: Identify nodes that differ from typical network patterns. An anomaly is not necessarily malicious; it may be a prominent researcher or influential paper.

The lecture emphasizes that network data is interdependent: connected nodes generally cannot be treated as independent observations. Consequently, ordinary data-sampling assumptions and techniques may not apply directly.

Speakers and Sources Featured

• Speaker: One lecturer, unnamed in the subtitles. No guest speakers are identified.
• Sources and examples mentioned: Online teaching materials (not specified); a book recommended by the lecturer (title not stated in the subtitles); research studies referred to without full citations, including discussion of the “six degrees” phenomenon and a study about terrorist networks; and Larry Page and Sergey Brin in connection with PageRank.
• Platforms and systems used as examples: WhatsApp, Facebook, Twitter, Flipkart, Amazon, Google, Bing, Yahoo, and the World Wide Web.