Are Claims Grounded in Data? An Empowering Linking Approach for Misalignment Identification in Online Data-Driven Discussions

Abstract

Textual content is effectively supported by data visualization in many contexts, allowing readers to assess the written content by looking at data visualizations. Among the fields that benefit from it are journalism and academia, from online blogs to analytical reports. While research has explored the effectiveness of linking text and data visualizations investigating learning, engagement, and improvement in readers’ skills regarding better information understanding and recall, we could commit more to using this practice in information verification. This paper investigates the impact of linking text modeled as discussions and data visualizations to identify text-to-visualization misalignment. First, we propose a text-visualization interaction-based approach, named HOOD, to aid users in online data-driven discussions by integrating deictic data visualization with messages, avoiding tab switching, and enabling fine-grain text-to-visualization linking. Then, we present a within-subjects design user study comparing HOOD with two alternative text-to-visualization layouts (also referred to as data access modalities) engaging 22 participants in identifying misalignment in text linked to data visualizations in 18 hand-crafted discussions. Our findings reveal that participants experienced 10% improvement with HOOD reaching up to 80% of success rate compared to alternative data access modalities. Furthermore, the time required for assessments remarkably decreased, and participants consistently rated HOOD as the preferred data access modality.

Keywords

• Data-driven discussions
• data visualization
• deixis
• linking
• misalignment
• user study