Let's dig into the data (from France)bookmark_border

In social science research, grounded theory (GT) is widely adopted to analyze and code textual data. While large language models (LLMs) open new possibilities for coding at scale, they complicate the GT workflow by generating excessive codes and introducing hallucinations, posing challenges for experts to audit and revise the coding results. To address these challenges, we conducted a two-year design study and derived a series of domain needs and design tasks via interviews with scientists in entrepreneurial management. Subsequently, we presented PROVA-GT, a visual analytics system facilitating the GT workflow by supporting both the forward process of code generation and the backward process of validation and refinement, which are interconnected and iterative. Specifically, our system features a provenance workflow that leverages the agentic LLM and novel visual designs to assist experts in the backward process. To evaluate the usefulness of PROVA-GT and its generalizability in different domains, we reported two in-depth expert studies on entrepreneurial inspirations and trust erosion, respectively. We also conducted a user study to assess the effectiveness of the agentic workflow.

Large Language Models (LLMs) exhibit representational and syntactic biases that are difficult to evaluate due to the stochastic nature of text generation. Standard auditing methods rely on a single output inspection or static automated metrics. These approaches obscure the underlying probability distributions and fail to capture biases hidden in lower-probability generation branches. This paper introduces TreeTracer, a visual analytics tool designed to evaluate LLM bias through aggregated comparison. Using a systematic perturbation analysis pipeline, the tool replaces ontology-defined terms in each input prompt, aggregates hundreds of stochastic generations into a syntax-aligned hierarchical structure, and then performs classification-aware node merging with an auxiliary language model. The resulting structure is visualized through a custom Sankey diagram. By juxtaposing two ontology-driven trees, the workspace enables direct comparison between semantic contexts and supports systematic bias detection. Because any visualization reflects only a subset of the model’s learned behavior, the system further applies contrastive inference to compute and directly display counterfactual token probabilities across contexts, reducing the risk of misinterpreting the presence of bias. We validate the workspace through case studies comparing an unaligned baseline model GPT-2 XL against the constitutionally aligned Apertus models. The visual aggregation successfully exposes hidden representational harms, such as counterfactual pronoun suppression and conversational marginalization of individuals. Our user and expert studies demonstrate that the aggregated comparative interface minimizes cognitive load and empowers analysts to effectively identify systemic biases.

Forestry presents complex, irregular environments that are underrepresented in the visualization literature, despite the increasing availability of large-scale LiDAR and multispectral datasets. Extracting meaningful information from this data requires analyzing subtle structural and physiological variations among individual trees, a task that remains challenging due to the irregular, multi-layered nature of forests. In this paper, we present a visualization-driven approach for the interactive exploration and analysis of 3D forestry point clouds at the tree level, enriched with Normalized Difference Vegetation Index (NDVI) metrics. Our web-based system, entitled ForestVis, enables users to navigate dense forest environments, query individual tree attributes, and analyze vegetation metrics within their spatial context. The system design is informed by a semi-structured interview with forestry domain experts, from which we derive a set of analytical tasks emphasizing tree-centric abstractions, interactive exploration, and scalability to large datasets.} Through illustrative use cases, we show how the system supports applications such as precision forestry, biodiversity monitoring, and detection of tree stress. More broadly, this work highlights the potential of interactive visualization providing novel tools and frameworks for ecological research and sustainable forest management. We evaluate ForestVis through two expert studies: a qualitative study of usability and analytical reasoning, and a quantitative study reporting objective task measures together with benchmarks of rendering, streaming, and interaction performance across two hardware configurations. Our evaluation suggested that our tool is effective in helping domain experts explore specific forest tasks, in large forest point cloud datasets, while maintaining computational performance. Our findings suggest that different point cloud rendering styles are suited to different tasks, whether analyzing individual trees or groups of trees, and interactive filtering helped experts relate vegetation indices to the point-cloud instance segmentation and to tree height.

Data visualizations on social media are often encountered as small thumbnails embedded in rapidly scrolling feeds. These entry-stage visualizations can elicit markedly different responses across users, yet little is known about how to systematically characterize such heterogeneity. We introduce attraction profiles, a user-level construct capturing consistent tendencies in first-exposure interactions with visualization thumbnails. To study these profiles, we construct a dataset of 1,349 thumbnails from authoritative news organizations and record click behaviors from 800 participants in a feed-style browsing study. Using this dataset, we develop the Attraction Profile Modeling Framework, which extracts visual and textual features and decouples topical influences to infer content-independent user representations. Based on these profiles, we introduce the Visualization Attraction Type Indicator (VATI), a structured typology that applies Exploratory Factor Analysis to uncover latent factors and maps them to five interpretable dimensions: Analytical Depth, Epistemic Drive, Visual Complexity Sensitivity, Figurative Tendency, and Structural Dependence. Each user is represented as a vector along these dimensions, enabling systematic comparison. Evaluation through ablation and user studies shows that APM captures intrinsic attraction patterns, while VATI provides an interpretable representation of multidimensional user tendencies.

Information visualization is inherently a field that brings together various research domains. Roughly speaking, we may identify two perspectives: the design perspective, revolving around how to ensure that a human can work effectively with the visual representations of data and the tools that offer them, and the algorithmic perspective, focusing on how to automatically create such visual representations. Munzner's model for visualization design places design choices before algorithmic considerations. It offers predominantly a design perspective; as a consequence, applications of this model may consider the algorithmic perspective as an afterthought, bypassing a step that translates the design into the formalism necessary for algorithmic study. As a result, the design may be entangled with the algorithms used to compute a visualization. Focusing on layout algorithms, we explore the ramifications of this entanglement: quality often goes undefined and unmeasured, and ad-hoc heuristics tend to be applied, reducing trustworthiness and potentially leading to incorrect conclusions. We look at how we may complement Munzner's model---the design perspective---with a clear model of the algorithmic perspective, to obtain a formal, measured understanding of the interplay between visualizations and the algorithms used to create them. Paradoxically, the solution lies in a clearer separation of concerns between design and algorithm. We argue that this model leads to better comparison between approaches, a more fine-grained understanding of their strengths and weaknesses, and allows for uncovering new opportunities, as to eventually lead to a better understanding of visualization.

The rise of AI agents introduces a fundamental shift in Visual Analytics (VA), in which agents act as a new user group. Current agentic approaches—based on computer vision and raw DOM access—fail to perform VA tasks accurately and efficiently. This paper introduces the Visual Analytics Context Protocol (VACP), a framework designed to make VA applications "agent-ready" that extends generic protocols by explicitly exposing application state, available interactions, and mechanisms for direct execution. To support our context protocol, we contribute a formal specification of AI agent requirements and knowledge representations in VA interfaces. We instantiate VACP as a library compatible with major visualization grammars and web frameworks, enabling augmentation of existing systems and the development of new ones. Our evaluation across a compelling set of VA tasks demonstrates that VACP-enabled agents achieve higher success rates in interface interpretation and execution compared to current agentic approaches, while reducing token consumption and completion time. VACP closes the gap between human-centric VA interfaces and machine perceivability, ensuring agents can reliably act as collaborative users in VA systems. More information under https://vacp.dev.

We introduce ProgressiVis, a language and graphical user interface designed to scale interactive visualization and visual analytics by relying on the Progressive Data Analysis (PDA) programming paradigm. Existing systems cannot remain interactive for open-ended exploration with large datasets and complex analyses because their processing time increases with size and complexity; at some point, time exceeds human latency limits (≈10 seconds). PDA systems guarantee that their latency remains within specified limits. They split long computations into smaller execution chunks so that processing a large dataset or applying complex computations still takes time, but they produce intermediary results continuously, improving quality and allowing analysts to monitor the computation, interact with it, and make decisions, often before the whole computation is complete. Implementing PDA systems is not currently supported by any programming environment, forcing programmers to reimplement the data analysis and visualization components from scratch, a long and difficult task. ProgressiVis is a language and system designed to run its programs progressively, including data management, data analysis, and visualizations. It introduces novel programming constructs and provides new services, such as progressive quality assessment, as well as JupyterLab notebook extensions to better support progressive visualization and program execution. This article describes the language, shows examples of ProgressiVis programs that perform visualization and analysis at scale, and explains how it addresses the technical and scientific issues raised by PDA, enabling data exploration to reach new limits.