Title: Sentiment and stance visualization of textual data for social media
Subject: Computer and information science
Faculty: Faculty of technology
Date: Monday 15 April 2019 at 9.15 am
Place: Room Wicksell, building K, Växjö
External reviewer: Associate professor Ross Maciejewski, Arizona State University, USA
Examining committee: Professor Jarke van Wijk, Eindhoven University of Technology, the Netherlands; professor Jörn Kohlhammer, Technische Universität Darmstadt, Germany; professor Gerik Scheuermann, Leipzig University, Germany
Chairperson: Associate professor Morgan Ericsson, Department of computer science and media technology, Linnaeus University
Supervisor: Professor Andreas Kerren, Department of computer science and media technology, Linnaeus University
Examiner: Professor Welf Löwe, Department of computer science and media technology, Linnaeus University
Spikning: Friday 15 March 2019 at 11.00 am at the University library in Växjö
Rapid progress in digital technologies has transformed the world in many ways during the past few decades, in particular, with the new means of communication such as social media. Social media platforms typically rely on textual data produced or shared by the users in multiple timestamped posts. Analyses of such data are challenging for traditional manual methods that are unable to scale up to the volume and the variety of the data. While computational methods can partially address these challenges, they have to be used together with the methods developed within information visualization and visual analytics to gain knowledge from the text data by using interactive visual representations.
One of the most interesting aspects of text data is related to expressions of sentiments and opinions. The corresponding task of sentiment analysis has been studied within computational linguistics, and sentiment visualization techniques exist as well. However, there are gaps in research on the related task of stance analysis, dedicated to subjectivity that is not expressible only in terms of sentiment. Research on stance is an area of interest in linguistics, but support by computational and visual methods has been limited so far. The challenges related to definition, analysis, and visualization of stance in textual data call for an interdisciplinary research effort. The StaViCTA project addressed these challenges with a focus on written text in English. The corresponding results in the area of visualization are reported in this work, based on multiple publications.
The main goal of this dissertation is to define, categorize, and implement means for visual analysis of sentiment and stance in textual data, in particular, for social media. Our work is based on the theoretical framework and automatic classifier of stance developed by our project collaborators, involving multiple non-exclusive stance categories such as certainty and prediction. We define a design space for sentiment and stance visualization techniques based on literature surveys. We discuss multiple visualization and visual analytics approaches developed by us to facilitate the underlying research on stance analysis, data collection and annotation, and visual analysis of sentiment and stance in real-world text data from several social media sources. The work described in this dissertation was carried out in cooperation with domain experts in linguistics and computational linguistics, and our approaches were validated with case studies, expert user reviews, and critical discussion. The results of this work open up further opportunities for research in text visualization and visual text analytics. The potential application areas are academic research, business intelligence, social media monitoring, and journalism.