Abstract
As Large Language Models (LLMs) automate text categorization, the primary challenge for readers has shifted from coding documents to effectively visualizing them. This study evaluates 10 color schemes for text annotation, finding that non-analogous (mixed-temperature) and yellow-inclusive schemes significantly enhance information-seeking speed and accuracy, whereas red-heavy schemes tend to hinder performance.
Introduction
Color coding is essential for academic and professional reading, allowing for the rapid clustering of information types such as "Methods" versus "Findings." While LLMs like GPT-4 can now annotate vast amounts of text instantaneously, a standardized "best practice" for color selection in dense textual environments remains elusive. Most existing research focuses on maps or data visualizations rather than paragraphs of text. This paper addresses how hue and color temperature (warm vs. cool) influence a reader's ability to locate specific information within highlighted text.
Methodology
The researchers conducted a controlled study with 32 participants tasked with locating specific labels (e.g., "Method") within scholarly abstracts.
Experimental Design
The study utilized 10 distinct color schemes derived from the Material UI system, using Red, Yellow, Green, and Blue as base colors. These schemes were categorized into four groups based on their chromatic relationships:
- Monochromatic: Shades of a single color.
- Dichromatic: Combinations of two base colors.
- Analogous: Hues that are visually similar (e.g., Green-Blue).
- Non-analogous: Contrasting hues (e.g., Yellow-Blue).
To ensure legibility, all schemes were standardized to a contrast ratio of approximately 5.55:1 against black text. Participants were given an 8-second time limit per query to simulate rapid information seeking.
Dataset
The study employed 706 abstracts from the arXiv HCI field, labeled using the CODA-19 framework (Background, Purpose, Method, Finding, Other). A final selection of 120 abstracts was used for the main task, ensuring each contained the four primary classes.
Key Insights
The Importance of Contrast
Non-analogous schemes outperformed analogous ones. When colors are too similar, visual boundaries blur, increasing cognitive load. Conversely, disparate hues (mixing warm and cool colors) help readers distinguish sections more efficiently.
The "Yellow Priority" vs. "Red Distraction"
Yellow-inclusive schemes emerged as the most effective, yielding the fastest response times and highest participant preference. This suggests a "yellow priority" effect where the color is perceived as prominent yet easy on the eyes. In contrast, red-inclusive schemes resulted in the slowest response times. Participants found red "distracting," likely due to its cultural association with errors or high-alert signals, which may interfere with neutral information seeking.
Results
Quantitative analysis confirmed these qualitative observations. Non-analogous schemes achieved a significantly higher accuracy of 98% compared to 95.7% for analogous schemes ($p=0.001$).
As shown in the performance scatter plot, yellow-inclusive schemes were the fastest overall (mean 4.120s), while red-inclusive schemes were the slowest (mean 4.350s). Participant preference peaked with the Yellow-Green scheme (4.58/5) and was lowest for pure Red (2.42/5).
Discussion and Limitations
While high-contrast colors facilitate rapid information location, their impact on deep reading comprehension remains to be studied. Furthermore, the findings are currently limited to scholarly abstracts and may vary across different document types like legal or creative texts. Future work should also address accessibility for visually impaired readers and explore more complex color appearance models to account for individual perceptual differences.
Summary updated