Statistical characteristics of comic panel viewing times

Hikaru Ikuta, Leslie Wöhler, Kiyoharu AizawaScientific 2023 · DOI

TL;DR

This paper investigates how people read comics by analyzing the time spent viewing individual panels. They found that viewing time is strongly correlated with the amount of text in a panel, but also that reading speed varies significantly between individuals. The study also reveals that comic panel viewing times follow a common heavy-tailed distribution, suggesting that while text length is a primary driver, some panels capture attention for reasons beyond just their text content.

What Problem Does This Paper Address?

Comics are a unique medium, blending text and images to tell stories. Understanding how humans process this bimodal information is crucial for fields ranging from human-computer interaction (HCI) to cognitive science. While previous research has looked at eye movements or viewing times for short comic strips, there's a need for a more in-depth analysis of reading behavior across entire comic volumes, especially considering the growing popularity of digital comics and new reading formats like vertical scrolling. This paper aims to quantify attention in comic reading by analyzing panel viewing times and to uncover the statistical characteristics of this behavior, particularly how text and image content influence it.

Methods

The researchers developed a custom panel-wise comic viewing interface. This interface presents each comic panel individually, allowing users to navigate sequentially using arrow keys. This setup mimics the experience of reading vertical comics, which are increasingly popular on mobile devices. They conducted a user study where 18 participants read 7 comic book volumes using this interface. During the study, the viewing time for each panel was meticulously recorded. To analyze the data, they focused on two main aspects: the relationship between panel viewing time and text length, and the influence of image content. Text length was measured by the character count of speech bubbles within each panel. For image dependency, they analyzed "short-text panels" (those with minimal to no text).

Datasets

The study utilized 7 comic book volumes from the Manga109 dataset. This dataset is a valuable resource for comic research, providing annotations for panel bounding boxes and text. The researchers specifically selected these 7 volumes to represent a diverse range of genres and panel layouts. They collected over 99,000 viewing time data points from 18 participants, which they plan to release publicly.

Key Insights and Ideas

  • Text is King (for viewing time): Even in a bimodal medium like comics, the amount of text in a panel is a significant predictor of how long a reader will spend on it. This suggests that for many readers, processing the textual information is a primary driver of attention.
  • Individual Reading Speeds Matter: While text length is a factor, the rate at which this text is processed varies considerably between individuals. This implies that any model predicting comic viewing times needs to account for a user's baseline reading speed.
  • A Universal "Long Tail": Regardless of individual differences or text length, the distribution of viewing times for comic panels consistently follows a heavy-tailed pattern. This means that while most panels are viewed for a relatively short duration, a small number of panels are viewed for exceptionally long periods. This hints at other factors, like visual interest or narrative importance, influencing attention.
  • Mimicking Modern Reading: The panel-wise interface, while altering the traditional page-reading experience, closely resembles the emerging trend of vertical comic readers. This makes the findings directly relevant to the design and user experience of future digital comic platforms.

Key Results

  • The average viewing time for a comic panel is directly proportional to the length of the text within its speech bubbles.
  • The correlation between text length and viewing time was found to be moderate to strong (Pearson's R median of 0.4911, Spearman's R median of 0.7188), with individual correlations varying significantly.
  • The distribution of viewing times across all participants and for various text lengths exhibits a heavy-tailed characteristic, observable in log-log plots.
  • Participants generally found the panel-wise interface understandable, with an average understandability score of over 3.5 on a 5-point Likert scale, suggesting the altered reading format did not significantly hinder comprehension.

Limitations and Future Work

The authors acknowledge several limitations. Firstly, their method of measuring viewing time relies on button presses to transition between panels, which doesn't capture fine-grained eye movements or internal cognitive processing. Secondly, the panel-wise presentation removes the two-dimensional layout information of original comic pages, which might influence reading behavior. They also note that their statistical analysis is "underpowered" due to the sample size of 18 participants, though they mitigated this by collecting a large amount of data per participant.

Future work proposed includes:

  • Investigating the impact of specific image content and narrative "turning points" on viewing times.
  • Conducting experiments where participants annotate content to better understand why certain panels are revisited or viewed for longer.
  • Developing models that can predict viewing times by incorporating individual reading speeds and analyzing image features more deeply.
  • Exploring the relationship between viewing time and panel image features using image processing techniques.

Summary updated

Abstract

Comics are a bimodal form of art involving a mixture of text and images. Since comics require a combination of various cognitive processes to comprehend their contents, the analysis of human comic reading behavior sheds light on how humans process such bimodal forms of media. In this paper, we particularly focus on the viewing times of each comic panel as a quantitative measure of attention, and analyze the statistical characteristics of the distributions of comic panel viewing times. We create a user interface that presents comics in a panel-wise manner, and measure the viewing times of each panel through a user study experiment. We collected data from 18 participants reading 7 comic book volumes resulting in over 99,000 viewing time data points, which will be released publicly. The results show that the average viewing times are proportional to the text length contained in the panel's speech bubbles, with a rate of proportion differing for each reader, despite the bimodal setting. Additionally, we find that the viewing time for all users follows a common heavy-tailed distribution.
BibTeX
@inproceedings{ikuta2023statistical,
  author = {Hikaru Ikuta and Leslie Wöhler and Kiyoharu Aizawa},
  title = {{Statistical characteristics of comic panel viewing times}},
  booktitle = {Scientific},
  year = {2023},
  doi = {10.1038/s41598-023-47120-w},
}
Figure