Multiscale Visual Analytics of Dynamic Multivariate Networks Derived from Text Streams

Social media has become a popular platform for individuals to express their opinions on various topics, such as politics and social events. However, the textual information exchanged on these platforms is never neutral. Instead, it is always embedded in a context of viewpoints, beliefs, evidence, certainty, and uncertainty. To extract meaningful insights from online discussions, we need a better understanding of the dynamics of written discourse and how it affects the spread of information. The interplay of written discourse between multiple actors can pose unsolved challenges for the (visual) analysis of discussions. Despite these challenges, it is important to derive insights from these text streams. This is where the visual analytics methods come into play.

The goal of our project is to create efficient visual analytics tools that can help researchers make sense of important patterns in large-scale social-media text streams. These text streams are comprised of utterances between various actors forming dynamic multivariate networks. Additionally, we plan to integrate state-of-the-art graph neural networks and/or language models with visual analytics to improve the credibility and trustworthiness of the human-in-the-loop process.

In more detail, we plan to address the following steps:

• Fundamental concepts, characteristics, and potential design spaces of complex discourse models as dynamic multivariate networks in combination with a thorough requirement and task analysis.
• Visual analytics of the challenging interchange between a number of big multivariate networks and their evolution over time.
• Validation by expert reviews and usability studies that are the basis of our empirical goals.

The project is funded by ELLIIT, a strategic research environment in information technology and mobile communications supported by the Swedish government.