Forecasting user roles in online communities
Edwin Tye · ePrints Soton (University of Southampton) · 2015
There is a growing interest in setting up and modeling of online social networks, as there are major incentives (popularity, monetary) for owners and managers to understand their own communities.This is especially true for communities that were set up by businesses because of the time and money invested into building and maintaining such online platforms.We take the approach that these online communities operate similarly to an offline environment such that members of these social networks can be classified into several (social) roles, each asserting a different impact on the community.This project focuses on forecasting the number of users in each of the roles of an online community.The forecasting model is split into two different parts.Part I models the movement of existing users between the roles, which was formulated as a linear difference equation upon time discretization.It is treated as an optimization problem where the objective function can take either the form of a least squares or a non-linear iterative update based on the difference equation, both with box and linear constraints.Part II predicts the number of new users joining and currently inactive users returning to the community.It is examined from a statistical point of view, where we postulate that the number of new users joining is akin to arrivals in a queue.In order to find the driving factor behind the numbers of new user joining, a series of models are explored using different independent variables.Models are built and tested for the two problems separately at first, then later combined to produce forecasts and their performances were accessed.