Author: Felix Scholz
Supervisor: Hannes Werthner, Co-Supervisor: Julia Neidhardt
Abstract
Online news forums provide a longstanding way for the exchange of opinions with other users. These often-large user bases consist of a cross-section of all people with various opinions and motivations. By classifying recurring types of behaviors common in such a context, an effort is made to provide an understanding about the composition of even large user bases. Additionally, long-time observations could provide insights into the composition and development over time. This thesis creates a classification model for users of an online news forum. The model is created by combining exploratory and statistical data analysis and attempts to explain recurring behaviors found in such an online community context. Such a model can be used to analyze how the user base of a large community develops and provides a quick overview of the composition of its users.In order to keep the model generally applicable, and not too tied to the provided data, the thesis proposes six roles for actively participating users and one additional role for non-participating readers. These active roles are Taciturn, Silent Voter, Regular, Conversationalist, Power User, and Celebrity. The inactive role is called Lurker. The model performance was tested for its predictive power and achieved a macro F1 score of 0.8632.Applying the model to a set of long-term data, provided by the online news forum of derStandard.at, role distribution over time was analyzed, showing a gradual trend towards higher activity of forum users. Additionally, co-occurrences of roles in the long term behavior of users and the frequency of role switches were measured in order to evaluate whether users have inherent roles or show signs of various roles, which could be dependent on time or context.