Unveiling Insights: Surfacing Fine-Grained Discourse Acts in Short Free-Form Public Input Text through Visual Analytics

Abstract

Exploratory analysis of public-generated short free-form online text often plays a critical role in facilitating decision-making across various domains. Particularly, in the civic domain, analyzing and understanding the thoughts, opinions, and comments shared by the public on social media or online engagement platforms on various civic issues are crucial for tracking consensus and making informed policy decisions. However, the public inputs are often short, redundant, unstructured, and full of nuances and ambiguity with a lack of clear boundary between positive and negative stances, which demands significant time and effort to analyze. Coupled with a lack of guidelines to visualize short free-form text, designing visualizations to show meaningful insights from public input while preserving the context and semantic values remains a challenging task. In this work, we explored discourse acts as a fine-grained categorization to characterize public input beyond positive or negative stances. Furthermore, we applied McDonald's greedy approximation of global inference to prioritize the relevance and negate the redundancy present in the text. We integrated these approaches with an interactive prototype called Matryona that employs visualization techniques to provide a contextual summary of unstructured public input and enables multilayered exploration from different angles by controlling the information content. Our initial evaluation of the prototype suggests Matryona's potential for reducing ambiguity and eliciting nuances present in the short free-form text by using discourse acts and accelerating the analysis process by ranking text comments.