| Summary: | In the past decade we have been privileged to witness the creation and revolution of social media on the World Wide Web. The abundance of content available on the web allows us to analyze the way people interact and the roles they play in a conversation on a large scale. One such role is influencer in the conversation. Detecting influence can be useful for successful advertisement strategies, detecting terrorist leaders and political campaigning.
We explore influence in discussion forums, weblogs, and micro-blogs using several components that have been found to be indicators of influence. Our components are author traits, agreement, claims, argumentation, persuasion, credibility, and certain dialog patterns. In the first portion of this thesis we describe each of our system components. Each of these components is motivated by social science through Robert Cialdini’s “Weapons of Influence” [Cialdini, 2007]. The weapons of influence are Reciprocation, Commitment and Consistency, Social Proof, Liking, Authority, and Scarcity. We then show the method and experiments for classifying each component.
In the second part of this thesis we classify influencers across five online genres and analyze which features are most indicative of influencers in each genre. The online genres we explore are Wikipedia Talk Pages, LiveJournal weblogs, Political Forum discussions, Create Debate debate discussions, and Twitter microblog conversations. First, we describe a rich suite of features that were generated using each of the system components. Then, we describe our experiments and results including using domain adaptation to exploit the data from multiple online genres. Finally, we also provide a detailed analysis of a single weapon of influence, social proof, and its impact in detecting influence in Wikipedia Talk Pages. This provides a single example of the usefulness of providing comprehensive components in the detection of influence.
The contributions of this thesis include a system for predicting who the influencers are in online discussion forums. We provide an evaluation of a rich set of features inspired by social science. In our system, each feature set used to detect influence is complex and computed by a system component. This allows us to provide a detailed analysis as to why the person was chosen as an influencer. We also provide a comparison of differences across several online discussion datasets and exploit the differences across the different genres to provide further improvements in influence detection.
|
|---|
