Predicting Information Flow through Blogspace 

Jack Hebert, Junior, Computer Science and Engineering

Mentor: Dan Weld, Computer Science and Engineering 

Users of web-log (or "blog") communities create social networks by posting opinions and links that they find interesting, thus sharing them with other users who read the posts. While each user has limited influence over others, one user's post may cause another to spread the idea by re-posting a link or related opinion. Influences at this 'user-to-user' level form the basis of viral-marketing, a marketing tactic of growing importance. The ability to predict when users will re-post (or, in the terminology of viral marketing, become 'infected') would allow a marketer to determine the optimal subset of users to receive advertising such that the largest influence is gained for the minimum cost. My work evaluates two published methods for modeling these infections in order to determine their actual predictive ability. The 'Linear Threshold Model' and the 'Independent Cascade Model' have previously been used to identify which users are most influential and from which source users are becoming infected, but their predictive abilities have not yet been validated in terms of real-world data. My work analyzes the accuracy of these models and studies the causes of 'infection' by incorporating additional information into the models.