Social Network Analysis

Social network analysis underpins modern AI systems: recommendation engines use centrality to rank items, fraud detection finds broker nodes that link criminal communities, and LLM-based agents navigate knowledge graphs as networks of concepts. You will use these ideas to understand how information spreads in agentic loops, how Graph Neural Networks learn from neighbors, and how PageRank ranks web pages. It unlocks concrete capabilities: measuring importance, detecting communities, and predicting outcomes from network position.

Think of a social network as a map of people and the connections between them. Each person is a node, each relationship is an edge. The key insight is that your position in this map can be more important than your individual attributes. Some people are important because they know many others; some because they connect two otherwise separate groups; some because they are a short path to everyone else. Social network analysis gives you tools to measure these different kinds of importance quantitatively. It also helps you find groups of people who are densely connected to each other, and to model how ideas or behaviors spread through the network. The central point is that the structure of connections matters as much as the nodes themselves.

Imagine a city's road network. The city has neighborhoods, and roads connect them. Some intersections are important because they have many roads leading to them; some are important because they are the only bridge between two districts; some are important because they are on the shortest route between two ends of the city. Now imagine traffic flowing through the roads. The flow depends on the intersections and the capacities of the roads. In a social network, people are the intersections and relationships are the roads. Centrality measures are like asking: which intersection is most important? There are different answers because 'important' can mean 'has the most roads', 'is the only bridge', or 'lies on the shortest paths'. Communities are like districts: groups of intersections with many roads within and few roads out. Brokerage is a single intersection that connects two districts—whoever controls that intersection controls the flow between them. Diffusion is how a new traffic pattern spreads, from one intersection to its neighbors. The analogy breaks down when you consider that in a social network, the roads themselves can change—people can make or break relationships—and the flow is not just physical but includes information, influence, and disease. Also, in a city, intersections are physical locations; in a social network, the same person can be in multiple places at once conceptually.

Social network analysis is the study of social structure through the graph of relationships, using measures of centrality to quantify importance, community detection to find densely connected groups, brokerage to identify connectors between groups, and diffusion to model how things spread across ties.

This concept is newly introduced in your library, but it draws on your notes for Graph Theory, where you've recorded adjacency lists and matrices, BFS, and Dijkstra's algorithm. The core vocabulary—nodes, edges, paths, connectivity—comes straight from those notes. You've also studied PageRank, which is itself a centrality measure, and the idea of a random walk that we use in diffusion. There is no prior knowledge of social science assumed; we will build the measures from arithmetic and simple graph ideas you already know. Later, this unlocks Graph Neural Networks, where the network structure becomes the input to a machine learning model.