Social Media Network Analysis

It depends on the question. Follower networks capture relatively stable audience structure; retweet networks capture amplification and endorsement; reply networks capture direct (often cross-cutting) interaction; mention networks capture attention and reference. The same conversation yields different networks under each definition, so choose the tie that matches your construct — diffusion suggests retweets, dialogue and disagreement suggest replies — and be explicit about the choice, since it shapes every downstream metric.

Himelboim and colleagues distinguish structures such as divided (two dense, separate clusters indicating polarization), unified (one cohesive cluster), fragmented (many disconnected small groups), and hub-and-spoke (broadcast from a central account). The macro-structure summarizes how information flows: a divided network signals polarized echo chambers, a unified one signals shared conversation, and hub-and-spoke signals top-down broadcasting. Reading the whole-network shape adds a layer of insight beyond individual centrality scores.

Bot and coordinated accounts can inflate certain nodes' centrality and create artificial clusters, distorting both micro and macro structure. Platform API rate limits, sampling, private accounts, and content deletion mean the collected network is usually incomplete, and the missingness is rarely random. Both issues bias results in ways that are hard to fully correct, so credible studies document collection methods, attempt bot detection, and interpret structural findings with appropriate caution about completeness.