A lot of social interaction data in social media environments can be collected as event streams. A typical entry (called an event) in such a data stream consists of at least three fields: There is a sender, a receiver and a time stamp. An exemplary part of an event stream is given in table 1.

Sender    Receiver    Timestamp
A    B    2010-11-01 12:00:00
B    A    2010-11-01 12:12:00
C    B    2010-11-01 12:44:00
D    B    2010-11-01 13:01:00
...    ...    ...
Table 1: An exemplary event stream

Assuming that these events define a network between actors that influences future actor decisions about events, it is possible to test how certain structures within this networks define the probability of how senders of events chose receivers. The fourth decision in table 1, for example, is assumed to be influenced by all previous decisions. A binary view on the network before the fourth event is given in figure 1. Each event in the event stream has established a directed tie representing recent activity.

 (see Word-Document)
Figure 1: Network before the fourth event

The fourth event includes the decision of actor D to chose B as recipient. D is assumed to chose B dependent on the structures they are both embedded in. If in a very long data stream a lot of structurally equivalent decisions were observed, one could probably find a statistically significant effect for communication with “popular” actors, meaning actors with a high in-degree. As can be seen in figure 1, B has the most ingoing ties of all possible receivers (A, B and  C). The influence of structures on communication choices can be evaluated with a stochastic model in which unknown parameters (that weight structures of interest) are optimized to get the maximum likelihood of the observed event stream.

In a one-mode environment, like a communication network, there are a lot of other interesting structures. It can, for example, be estimated whether there is tendency for repeated using of the same communication partners, for reciprocal communication or for triadic effects like circles or transitive triangles.

In social media environments, however, often several different event driven networks can be measured at the same time. On web sites like facebook, for example, actors write messages (communication events), people establich frienship ties with others (friendship events) and people affiliate to products or parties (entity affiliation events). These different events types create several networks. Note, that in this context, also entities like products can be receivers of events, so there is also the possibility of two-mode networks.
 
The mentioned statistical model can be extended by including multi-network structures that represent whether, for example, affiliation to objects change the probability for communication, whether previous communication has an effect on the probability to establish friendship ties or whether friends tend to affiliate with the same entities.