Social Network Analysis Syllabus (Advanced Topic)
Course Organization
This course develops social network analysis (SNA) as both a set of quantitative methods and a lens for studying adversarial behavior, information spread, and influence in complex systems. Lectures progress from foundational concepts illustrated through animal behavior, through core computational methods, into applied analysis of information operations, and finally into simulation-based modeling of social dynamics.
For a broader overview of the Trust and Safety space, see the Trust & Safety class.
Lectures and Delivery Order
Lecture 1: Animal Sociality and SNA Fundamentals
- Source: sna_animal_networks
Uses animal social networks as a politically neutral entry point to introduce core SNA vocabulary: nodes, edges, weighted networks, directed vs. undirected graphs, assortative mixing, and homophily ("birds of a feather flock together"). Case studies draw from marmot fieldwork and broader ethology literature. Establishes foundational intuitions before applying methods to human social media networks.
Lecture 2: Animal Network Robustness and Node Removal
- Source: sna_animal_network_robustness
Explores what happens when individuals are removed from a network — strategically or randomly — using both animal and information network examples. Introduces node-level metrics (degree, betweenness centrality) and network-level metrics (connectedness, fragmentation). Applies these concepts to misinformation source rankings, illustrating how targeted interventions affect information flow. Lays groundwork for intervention analysis in later lectures.
Lecture 3: Community Detection
- Source: sna_community_detection
Core methodology lecture. Covers clustering coefficient, modularity maximization, the Louvain algorithm, and CONCOR (convergence of iterated correlations). Framed around a "Locate Groups" report assignment. Students learn to identify cohesive subgroups and evaluate the quality of detected communities using modularity scores.
Lecture 4: Stance Detection via Label Propagation
- Source: sna_stance_detection
Applied method that builds directly on community structure from Lecture 3. Introduces stance detection using hashtag-seeded label propagation over retweet networks. Covers how stance labels spread from users to hashtags and back, general label propagation algorithms, confidence calibration, and the choice between propagation strategies. Optional extension covers text-based stance detection.
Lecture 5: Information Operations
- Source: sna_information_operations
Conceptual overview of adversarial social behavior using the BEND framework (Boost, Engage, Neutralize, Distort). Connects community structure and stance to coordinated inauthentic behavior. Discusses dynamic multi-agent scenarios in which adversarial actors attempt to shift population-level stance. Sets up Lecture 6's detection approach.
Lecture 6: Information Operations Detection
- Source: sna_information_operations_detection
Technical case study of detection using paid link schemes and SEO manipulation as the adversarial domain. Covers how to identify coordinated link schemes, distinguish paid from organic linking, and use LLMs to label the political bias of news sites at scale (case study: Iranian news network). Draws on SEO network construction and classification methods introduced in Lectures 2 and 4.
Lecture 7: Social Influence Modeling
- Source: sna_social_influence_modeling
Introduces agent-based modeling (ABM) as a complement to network analysis. Develops a co-evolutionary stance-influence model where network structure and agent opinions update simultaneously. Key findings: minority stances exhibit tipping points around 25% adoption; optimal confederates target local ego-networks rather than global hubs. Discusses validation against real data and recovery from polarized states.
Lecture 8: Information Diffusion and Population Modeling
- Source: sna_population_modeling
Broadest lens in the course. Applies epidemiological-style population models (SEIRM, Friedkin social influence model) to information spread and opinion dynamics. Multi-agent scenarios allow virtual experiments: given an observed information environment, what policies lead a population toward a desired trajectory? LLMs are introduced as tools for constructing action distributions and translating between the agent-level model and the real information environment.