model behavior visualization

Decision support in visualization research lacks systematic characterization of decision-context frameworks, and existing task models fail to guide design for real-world decision scenarios. Method: We propose a structured decision-problem analysis framework that, for the first time, decomposes decision problems into three core attributes—data, user, and task context—and explicitly articulates their constraints and implications for visual encoding and interaction design. Grounded in task-modeling principles and visualization design theory, we develop an operational feature-description system for decision problems and validate it through multi-case analysis. Contribution/Results: The framework addresses the critical gap in traditional task models—neglect of decision context—and provides the first systematic theoretical tool for decision-oriented visualization research. It reveals limitations in current design practices and identifies concrete pathways to enhance decision-support efficacy in authentic settings.

Existing visual analytics (VA) process models inadequately capture the dynamic roles and bidirectional interactions among humans, data, and models—particularly regarding explainable AI (XAI), knowledge externalization, and closed-loop feedback. To address this gap, we propose HDMI Canvas: the first VA framework centered on triadic, bidirectional human–data–model interaction, integrating descriptive and generative capabilities. Grounded in human factors engineering, XAI principles, knowledge externalization theory, and closed-loop feedback design, HDMI Canvas emphasizes co-evolution between human cognition and model outputs, enabling interdisciplinary collaboration and stakeholder engagement. Evaluated through two empirical case studies, the framework significantly enhances the systematicity of VA process design, the effectiveness of user-centered practices, and the interpretability of cross-domain communication. HDMI Canvas thus advances a new paradigm for human–AI symbiotic intelligent analytics. (136 words)

To address the dual challenges of insufficient personalization and low efficiency in interactive exploration for automated insight discovery, this paper proposes InsightMap—a map-metaphor-based framework for insight visualization and hybrid discovery. Methodologically, it formalizes data insights as measurable, layout-aware data objects; introduces a similarity metric integrating semantic and statistical features; and establishes a hybrid paradigm that synergistically combines automated mining with interactive exploration. InsightMap enables seamless transitions from global overviews to localized deep-dive analysis. Through multiple case studies and user experiments, InsightMap reduces average task completion time by 37% and achieves a user satisfaction score of 4.8/5.0, demonstrating significant improvements in both insight discovery efficiency and personalized adaptability.

Current heat-risk assessments rely on numerical models, which suffer from low spatiotemporal resolution and inability to capture dynamic couplings among environmental, social, and behavioral factors—limiting actionable risk-informed decision-making. To address this, we propose *Havior*, the first integrated framework combining numerical simulation with news semantics for heat-risk visualization and analysis. It introduces two novel glyph paradigms: *thermoglyphs* for thermal dynamics and *news glyphs* for event semantics; employs an expert-guided, LLM-driven semantic extraction method for interpretable, dynamic risk-factor modeling; and unifies heterogeneous multi-source data within an interactive visualization interface. Evaluated on the 2022 China heatwave, Havior achieved a 92.3% F1-score in news-based risk information extraction. Six domain experts unanimously affirmed its substantial improvement in depth of risk perception and operational decision support.

This study investigates the discrepancies between large language models (LLMs) and human cognition in high-level chart understanding, with a focus on interpreting designer intent and extracting complex data patterns. Through qualitative user studies, it systematically compares the higher-order interpretation strategies employed by humans and LLMs on line charts, bar charts, and scatter plots, while analyzing LLM outputs and reasoning pathways under three distinct prompting conditions. The work reveals, for the first time, that LLMs consistently adopt a structured enumeration strategy rather than constructing coherent trend-based narratives, and their explanatory patterns remain remarkably stable across different prompts. In contrast, humans demonstrate a superior ability to synthesize holistic, narrative-driven interpretations. These findings highlight fundamental mechanistic limitations of LLMs in visual reasoning and offer critical insights for future model design.

This work addresses the prevalence of misleading visualizations resulting from violations of fundamental design principles, a problem exacerbated by existing tools’ lack of contextual understanding and the unreliability of general-purpose large language models in providing actionable guidance. To bridge this gap, the authors propose a novel approach that integrates chart de-rendering, vision-language reasoning, and a knowledge base of visualization principles to reconstruct structured chart representations from images, identify design flaws, and generate interpretable, executable improvement suggestions. The system enables human-in-the-loop interactive optimization and re-rendering. Evaluated on 1,000 charts from the Chart2Code benchmark, it produced 10,452 recommendations, clustered into 10 categories—including axis formatting and color accessibility—demonstrating its effectiveness in enhancing both visualization quality and user literacy.

This study investigates the mechanisms and pathways through which human domain knowledge is integrated into machine learning (ML) workflows via visual analytics. Building upon a systematic review of over 200 VIS4ML papers, the authors develop a coding framework encompassing four dimensions: machine learning, visualization, interaction, and action. By synthesizing perspectives from model construction and information-theoretic cost–benefit analysis, they propose the first unified explanatory framework for knowledge injection in ML. The work elucidates the pivotal role of interactive visualization in optimizing ML workflows and systematically maps the multidimensional pathways through which human expertise is incorporated. This contribution provides both theoretical grounding and empirical foundations for advancing research and practice in the VIS4ML community.

This study addresses the cognitive bias in scatterplots where “data-induced grouping”—arising from the interplay between data values and visual encoding—leads users to misinterpret spatial arrangements as meaningful patterns. Through two user studies, the authors systematically demonstrate the prevalence of this phenomenon, develop the first perceptual model capable of predicting whether users perceive a given set of points as a coherent group, and propose a visualization intervention strategy that integrates user perception with data reordering. Notably, the model effectively captures users’ tendency to group points based on trends even in nominal data contexts. Applied to visualization diagnosis and optimization, this approach significantly enhances the accuracy and reliability of graphical representations.