Eytan Adar

John Joon Young Chung, Eytan Adar

While diffusion-based text-to-image (T2I) models provide a simple and powerful way to generate images, guiding this generation remains a challenge. For concepts that are difficult to describe through language, users may struggle to create prompts. Moreover, many of these models are built as end-to-end systems, lacking support for iterative shaping of the image. In response, we introduce PromptPaint, which combines T2I generation with interactions that model how we use colored paints. PromptPaint allows users to go beyond language to mix prompts that express challenging concepts. Just as we iteratively tune colors through layered placements of paint on a physical canvas, PromptPaint similarly allows users to apply different prompts to different canvas areas and times of the generative process. Through a set of studies, we characterize different approaches for mixing prompts, design trade-offs, and socio-technical challenges for generative models. With PromptPaint we provide insight into future steerable generative tools.

Harmanpreet Kaur, Eytan Adar, Eric Gilbert et al.

Understanding how ML models work is a prerequisite for responsibly designing, deploying, and using ML-based systems. With interpretability approaches, ML can now offer explanations for its outputs to aid human understanding. Though these approaches rely on guidelines for how humans explain things to each other, they ultimately solve for improving the artifact -- an explanation. In this paper, we propose an alternate framework for interpretability grounded in Weick's sensemaking theory, which focuses on who the explanation is intended for. Recent work has advocated for the importance of understanding stakeholders' needs -- we build on this by providing concrete properties (e.g., identity, social context, environmental cues, etc.) that shape human understanding. We use an application of sensemaking in organizations as a template for discussing design guidelines for Sensible AI, AI that factors in the nuances of human cognition when trying to explain itself.

Joshua Ashkinaze, Ruijia Guan, Laura Kurek et al.

Large language models (LLMs) are trained on broad corpora and then used in communities with specialized norms. Is providing LLMs with community rules enough for models to follow these norms? We evaluate LLMs' capacity to detect (Task 1) and correct (Task 2) biased Wikipedia edits according to Wikipedia's Neutral Point of View (NPOV) policy. LLMs struggled with bias detection, achieving only 64% accuracy on a balanced dataset. Models exhibited contrasting biases (some under- and others over-predicted bias), suggesting distinct priors about neutrality. LLMs performed better at generation, removing 79% of words removed by Wikipedia editors. However, LLMs made additional changes beyond Wikipedia editors' simpler neutralizations, resulting in high-recall but low-precision editing. Interestingly, crowdworkers rated AI rewrites as more neutral (70%) and fluent (61%) than Wikipedia-editor rewrites. Qualitative analysis found LLMs sometimes applied NPOV more comprehensively than Wikipedia editors but often made extraneous non-NPOV-related changes (such as grammar). LLMs may apply rules in ways that resonate with the public but diverge from community experts. While potentially effective for generation, LLMs may reduce editor agency and increase moderation workload (e.g., verifying additions). Even when rules are easy to articulate, having LLMs apply them like community members may still be difficult.

Tenghao Ji, Eytan Adar

A challenge in fine-tuning text-to-image diffusion models for specific topics is to select good examples. Fine-tuning from image sets of varying quality, such as Wikipedia Commons, will often produce poor output. However, training images that \textit{do} exemplify the target concept (e.g., a \textit{female Mountain Bluebird}) help ensure that the generated images are similarly representative (e.g., have the prototypical blue-wings and gray chest). In this work, we propose QZLoRA, a framework to select images for low-rank adaptation (LoRA). The approach leverages QuizRank, a method to automatically rank images by treating them as an `educational intervention' and `quizzing' a VLM. We demonstrate that QZLoRA can produce better aligned, photorealistic images with fewer samples. We also show that these fine-tuned models can produce stylized that are similarly representative (i.e., illustrations). Our results highlight the promise of combining automated visual reasoning with parameter-efficient fine-tuning for topic-adaptive generative modeling.

Taewook Kim, Hyomin Han, Eytan Adar et al.

Generative AI has the potential to create a new form of interactive media: AI-bridged creative language arts (CLA), which bridge the author and audience by personalizing the author's vision to the audience's context and taste at scale. However, it is unclear what the authors' values and attitudes would be regarding AI-bridged CLA. To identify these values and attitudes, we conducted an interview study with 18 authors across eight genres (e.g., poetry, comics) by presenting speculative but realistic AI-bridged CLA scenarios. We identified three benefits derived from the dynamics between author, artifact, and audience: those that 1) authors get from the process, 2) audiences get from the artifact, and 3) authors get from the audience. We found how AI-bridged CLA would either promote or reduce these benefits, along with authors' concerns. We hope our investigation hints at how AI can provide intriguing experiences to CLA audiences while promoting authors' values.

Chen Liang, Xirui Jiang, Naihao Deng et al.

AI agents operating on user interfaces must understand how interfaces communicate state and feedback to act reliably. As a core communicative modality, animations are increasingly used in modern interfaces, serving critical functional purposes beyond mere aesthetics. Thus, understanding UI animation is essential for comprehensive interface interpretation. However, recent studies of Vision Language Models (VLMs) for UI understanding have focused primarily on static screenshots, leaving it unclear how well these models handle dynamic UI animations. To address this gap, we created AniMINT, a novel dataset of 300 densely annotated UI animation videos. We systematically evaluate state-of-the-art VLMs on UI animation understanding, including their abilities to perceive the animation effects, identify animation purposes, and interpret animation meaning. Our results show that VLMs can reliably detect primitive motion. However, their high-level animation interpretation remains inconsistent, with substantial gaps relative to human performance. Finally, we use Motion, Context, and Perceptual Cues (MCPC) to probe factors affecting VLM performance, revealing key bottlenecks and directions for future improvement.

Elsie Lee-Robbins, Eytan Adar

Using learning objectives to define designer intents for communicative visualizations can be a powerful design tool. Cognitive and affective objectives are concrete and specific, which can be translated to assessments when creating, evaluating, or comparing visualization ideas. However, while there are many well-validated assessments for cognitive objectives, affective objectives are uniquely challenging. It is easy to see if a visualization helps someone remember the number of patients in a clinic, but harder to observe the change in their attitudes around donations to a crisis. In this work, we define a set of criteria for selecting assessments--from education, advocacy, economics, health, and psychology--that align with affective objectives. We illustrate the use of the framework in a complex affective design task that combines personal narratives and visualizations. Our chosen assessments allow us to evaluate different designs in the context of our objectives and competing psychological theories.

Xingjian Zhang, Tianhong Gao, Suliang Jin et al.

Large language models (LLMs) are increasingly used as raters for evaluation tasks. However, their reliability is often limited for subjective tasks, when human judgments involve subtle reasoning beyond annotation labels. Thinking traces, the reasoning behind a judgment, are highly informative but challenging to collect and curate. We present a human-LLM collaborative framework to infer thinking traces from label-only annotations. The proposed framework uses a simple and effective rejection sampling method to reconstruct these traces at scale. These inferred thinking traces are applied to two complementary tasks: (1) fine-tuning open LLM raters; and (2) synthesizing clearer annotation guidelines for proprietary LLM raters. Across multiple datasets, our methods lead to significantly improved LLM-human agreement. Additionally, the refined annotation guidelines increase agreement among different LLM models. These results suggest that LLMs can serve as practical proxies for otherwise unrevealed human thinking traces, enabling label-only corpora to be extended into thinking-trace-augmented resources that enhance the reliability of LLM raters.

Tenghao Ji, Eytan Adar

Images play a vital role in improving the readability and comprehension of Wikipedia articles by serving as `illustrative aids.' However, not all images are equally effective and not all Wikipedia editors are trained in their selection. We propose QuizRank, a novel method of image selection that leverages large language models (LLMs) and vision language models (VLMs) to rank images as learning interventions. Our approach transforms textual descriptions of the article's subject into multiple-choice questions about important visual characteristics of the concept. We utilize these questions to quiz the VLM: the better an image can help answer questions, the higher it is ranked. To further improve discrimination between visually similar items, we introduce a Contrastive QuizRank that leverages differences in the features of target (e.g., a Western Bluebird) and distractor concepts (e.g., Mountain Bluebird) to generate questions. We demonstrate the potential of VLMs as effective visual evaluators by showing a high congruence with human quiz-takers and an effective discriminative ranking of images.

Yoonjoo Lee, Kihoon Son, Tae Soo Kim et al.

As Large Language Models (LLMs) are nondeterministic, the same input can generate different outputs, some of which may be incorrect or hallucinated. If run again, the LLM may correct itself and produce the correct answer. Unfortunately, most LLM-powered systems resort to single results which, correct or not, users accept. Having the LLM produce multiple outputs may help identify disagreements or alternatives. However, it is not obvious how the user will interpret conflicts or inconsistencies. To this end, we investigate how users perceive the AI model and comprehend the generated information when they receive multiple, potentially inconsistent, outputs. Through a preliminary study, we identified five types of output inconsistencies. Based on these categories, we conducted a study (N=252) in which participants were given one or more LLM-generated passages to an information-seeking question. We found that inconsistency within multiple LLM-generated outputs lowered the participants' perceived AI capacity, while also increasing their comprehension of the given information. Specifically, we observed that this positive effect of inconsistencies was most significant for participants who read two passages, compared to those who read three. Based on these findings, we present design implications that, instead of regarding LLM output inconsistencies as a drawback, we can reveal the potential inconsistencies to transparently indicate the limitations of these models and promote critical LLM usage.

Dongping Zhang, Eytan Adar, Jessica Hullman

Probabilistic graphs are challenging to visualize using the traditional node-link diagram. Encoding edge probability using visual variables like width or fuzziness makes it difficult for users of static network visualizations to estimate network statistics like densities, isolates, path lengths, or clustering under uncertainty. We introduce Network Hypothetical Outcome Plots (NetHOPs), a visualization technique that animates a sequence of network realizations sampled from a network distribution defined by probabilistic edges. NetHOPs employ an aggregation and anchoring algorithm used in dynamic and longitudinal graph drawing to parameterize layout stability for uncertainty estimation. We present a community matching algorithm to enable visualizing the uncertainty of cluster membership and community occurrence. We describe the results of a study in which 51 network experts used NetHOPs to complete a set of common visual analysis tasks and reported how they perceived network structures and properties subject to uncertainty. Participants' estimates fell, on average, within 11% of the ground truth statistics, suggesting NetHOPs can be a reasonable approach for enabling network analysts to reason about multiple properties under uncertainty. Participants appeared to articulate the distribution of network statistics slightly more accurately when they could manipulate the layout anchoring and the animation speed. Based on these findings, we synthesize design recommendations for developing and using animated visualizations for probabilistic networks.

Elsie Lee-Robbins, Shiqing He, Eytan Adar

Despite the ubiquity of communicative visualizations, specifying communicative intent during design is ad hoc. Whether we are selecting from a set of visualizations, commissioning someone to produce them, or creating them ourselves, an effective way of specifying intent can help guide this process. Ideally, we would have a concise and shared specification language. In previous work, we have argued that communicative intents can be viewed as a learning/assessment problem (i.e., what should the reader learn and what test should they do well on). Learning-based specification formats are linked (e.g., assessments are derived from objectives) but some may more effectively specify communicative intent. Through a large-scale experiment, we studied three specification types: learning objectives, insights, and assessments. Participants, guided by one of these specifications, rated their preferences for a set of visualization designs. Then, we evaluated the set of visualization designs to assess which specification led participants to prefer the most effective visualizations. We find that while all specification types have benefits over no-specification, each format has its own advantages. Our results show that learning objective-based specifications helped participants the most in visualization selection. We also identify situations in which specifications may be insufficient and assessments are vital.

Hariharan Subramonyam, Colleen Seifert, Eytan Adar

Thinking of technology as a design material is appealing. It encourages designers to explore the material's properties to understand its capabilities and limitations, a prerequisite to generative design thinking. However, as a material, AI resists this approach because its properties emerge as part of the design process itself. Therefore, designers and AI engineers must collaborate in new ways to create both the material and its application experience. We investigate the co-creation process through a design study with 10 pairs of designers and engineers. We find that design 'probes' with user data are a useful tool in defining AI materials. Through data probes, designers construct designerly representations of the envisioned AI experience (AIX) to identify desirable AI characteristics. Data probes facilitate divergent thinking, material testing, and design validation. Based on our findings, we propose a process model for co-creating AIX and offer design considerations for incorporating data probes in design tools.

Eytan Adar, Elsie Lee

Significant research has provided robust task and evaluation languages for the analysis of exploratory visualizations. Unfortunately, these taxonomies fail when applied to communicative visualizations. Instead, designers often resort to evaluating communicative visualizations from the cognitive efficiency perspective: "can the recipient accurately decode my message/insight?" However, designers are unlikely to be satisfied if the message went 'in one ear and out the other.' The consequence of this inconsistency is that it is difficult to design or select between competing options in a principled way. The problem we address is the fundamental mismatch between how designers want to describe their intent, and the language they have. We argue that visualization designers can address this limitation through a learning lens: that the recipient is a student and the designer a teacher. By using learning objectives, designers can better define, assess, and compare communicative visualizations. We illustrate how the learning-based approach provides a framework for understanding a wide array of communicative goals. To understand how the framework can be applied (and its limitations), we surveyed and interviewed members of the Data Visualization Society using their own visualizations as a probe. Through this study we identified the broad range of objectives in communicative visualizations and the prevalence of certain objective types.

Xin Rong, Joshua Luckson, Eytan Adar

Tuning machine learning models, particularly deep learning architectures, is a complex process. Automated hyperparameter tuning algorithms often depend on specific optimization metrics. However, in many situations, a developer trades one metric against another: accuracy versus overfitting, precision versus recall, smaller models and accuracy, etc. With deep learning, not only are the model's representations opaque, the model's behavior when parameters "knobs" are changed may also be unpredictable. Thus, picking the "best" model often requires time-consuming model comparison. In this work, we introduce LAMVI-2, a visual analytics system to support a developer in comparing hyperparameter settings and outcomes. By focusing on word-embedding models ("deep learning for text") we integrate views to compare both high-level statistics as well as internal model behaviors (e.g., comparing word 'distances'). We demonstrate how developers can work with LAMVI-2 to more quickly and accurately narrow down an appropriate and effective application-specific model.

Guy D. Rosin, Eytan Adar, Kira Radinsky

Search systems are often focused on providing relevant results for the "now", assuming both corpora and user needs that focus on the present. However, many corpora today reflect significant longitudinal collections ranging from 20 years of the Web to hundreds of years of digitized newspapers and books. Understanding the temporal intent of the user and retrieving the most relevant historical content has become a significant challenge. Common search features, such as query expansion, leverage the relationship between terms but cannot function well across all times when relationships vary temporally. In this work, we introduce a temporal relationship model that is extracted from longitudinal data collections. The model supports the task of identifying, given two words, when they relate to each other. We present an algorithmic framework for this task and show its application for the task of query expansion, achieving high gain.

Lada A. Adamic, Thomas M. Lento, Eytan Adar et al.

Social networks readily transmit information, albeit with less than perfect fidelity. We present a large-scale measurement of this imperfect information copying mechanism by examining the dissemination and evolution of thousands of memes, collectively replicated hundreds of millions of times in the online social network Facebook. The information undergoes an evolutionary process that exhibits several regularities. A meme's mutation rate characterizes the population distribution of its variants, in accordance with the Yule process. Variants further apart in the diffusion cascade have greater edit distance, as would be expected in an iterative, imperfect replication process. Some text sequences can confer a replicative advantage; these sequences are abundant and transfer "laterally" between different memes. Subpopulations of the social network can preferentially transmit a specific variant of a meme if the variant matches their beliefs or culture. Understanding the mechanism driving change in diffusing information has important implications for how we interpret and harness the information that reaches us through our social networks.