Predicting Cascades on Social Media
http://snap.stanford.edu/class/cs224w-readings/bakshy11influencers.pdf
In this paper, the authors investigate quantifying influence on twitter by tracking cascades. One important finding of the study was that predictions of which particular user or URL will generate large cascades are relatively unreliable. They discuss a concept of word-of-mouth diffusion. Word-of-mouth diffusion, as it sounds is the process of spreading the influence of information or a new product with certain special individuals who exhibit some desirable attributes such as credibility, expertise, enthusiasm, or centrality. Historically, studies of word-of-mouth diffusion have been unreliable because the network is generally un-observable and the observational data on diffusion are heavily biased towards “successful” diffusion events. However, with the advent of social media, the micro-blogging service Twitter is an ideal place to study diffusion processes. The main attributes that the study tracks are follower relationships and URLs shared. The specific process that the researchers observe is the influencers ability to post URLs which diffuse through the Twitter follower graph. They focused on “seed” influencers, who are people who post original content, not received through the follower graph. They quantify the influence of a given post simply by the number of users who repost the URL. Using this information combined with the individual’s attributes and past activity, the researchers try to predict the influence of that user. The purpose of a method like this is to gain insight into how to identify influencers ahead of time. In contrast, most marketing techniques only identify influencers in retrospect (using an ex-post explanation) while the study emphasizes the ex-ante prediction of influencers.