Analyzing user discussion dynamics in social media platforms

Abstract

Understanding how users behave when they connect to social networking sites creates opportunities for better interface design, more productive studies of social interactions, and improved design of content distribution systems. Traditionally, user behavior characterization methods, based on individual features of users, are not appropriate for online networking sites. In these environments, users interact with the site and with other users through a series of multiple interfaces that let them upload and view content, choose friends, rank favorite content, subscribe to users and allow many other interactions. Different types of interactions can be observed among different types of users, and interactions on the same topic between a set of users can have different patterns as well. The motivation is to explore user behavior and the underlying conversation patterns. How a set of users react to a piece of particular news, or for a political leader how a single user changes the polarity with time or how it is constant throughout the time which can be observed in his expressive online statements (e.g., - tweets, Reddit discussion, Facebook comments, etc.). We study user behavior and its patterns in primarily three parts. (i) We present a novel quantification of conflict in an online discussion. Our measure of conflict dynamics is continuous-valued, which we validate with manually annotated ratings. Firstly, we predict the probable degree of conflict a news article will face from its audience. We employ multiple machine learning frameworks for this task using various features extracted from news articles. Secondly, given a pair of users and their interaction history, we predict if their future engagement will result in a conflict. (ii) We study how the sentiment of a user towards entities can be predicted using the tweets the user has posted so far. We consider all the available sentiments of the entities for the prediction task. (iii) In this task by using the same twitter data, we study how the sentiment of a user to an entity changes with respect to time - with time how a user changes the polarity, whether it remains same or it varies after a certain point, if so how many users had the same kind of polarity drift for an entity.