A central problem in the analysis of observational data is inferring causal relationships - what are the underlying causes of the observed behaviors? With the recent proliferation of Big Data from online social networks, it has become important to determine to what extent social influence causes certain messages to ‘go viral’, and to what extent other causes also play a role. In this paper, we present a causal framework showing that social influence is confounded with personal similarity, traits of the focal item, and external circumstances. Combined with a set of qualitative considerations on the combination of these sources of causation, we show how this framework can enable investigators to systematically evaluate, strengthen and qualify causal claims about social influence, and we demonstrate its usefulness and versatility by applying it to a variety of common online social datasets.
|Name||Lecture Notes in Computer Science|