How Visual Clutter Affects You? : a study of Danmaku Commentary and Its Effect on Viewers’ Information Gain

Master’s Thesis:

How Visual Clutter Affects You?

A study of Danmaku Commentary and Its Effect on Viewers’

Information Gain

Student: Yanan Chen (11595582)

Supervisor: Hande Sungur

Graduate School of Communication

Master’s programme Communication Science

University of Amsterdam

Data of Completion: 31/01/2019

Abstract

Danmaku commentary as an emerging technology can enhance the original video content by synchronizing various user comments to the video time. From the perspective of media richness and cognitive overload, this study focuses on the informative feature of Danmaku and raises an important question of how Dammaku presentation density affect viewers’ information gain. An online experiment was conducted and a total of 112 Mandarin-speaking young adults have participated in this study. Results shows that higher levels of Danmaku density decreased viewers’ information gain from the video content. Specifically, participants showed lower memory performance and lower levels of knowledge enhancement when watching a Danmaku video with high density presentation. Moreover, higher levels of Danmaku density also distract more attention from video-viewing compared to lower levels of density. However, Danmaku density did not indirectly affect individuals’ information gain through the mediating cause of the perceived distraction. Additionally, the possible

moderating effect of previous Danmaku experience on distraction levels was examined, however, the results turn out that one’s experience with Danmaku did not interfere with the relationship between Danmaku presentation density and distraction. Potential limitations and directions for future studies are discussed.

How Visual Clutter Affects You? A study of Danmaku Commentary and Its Effect on Viewers’ Information Gain

With the advancement of the internet, individuals’ media consumption habits have been dramatically changed over recent years. According to the findings of the survey

conducted by Pew Research Center (September 2017), 61% of American young adults (18-29 years old) prefer to use online streaming services to watch television. In China, a latest report from China Netcasting Services Association shows that the number of online video viewers has reached 6.09 billion as of June 2018 (CNSA, November 2018, p. 5), of which, around 5.78 billion users watch videos on smartphones (p. 6). Many online video sites, like YouTube, provide a separate comment section below the video so that viewers can leave their post-watching reflections and give a like, dislike or reply to others’ comments on a video. Commentary system turns internet videos into a medium of communication. Recently, one innovative technology has created a more engaging watching context for viewers (Liu, Suh, & Wagner, 2016), which is called the Danmaku commentary system.

Danmaku, originated in Japan, is a real-time commentary system that enables viewers to send comments onto the screen during watching a video. This creates an immersive co-viewing experience for viewers with the features of immediacy, interactivity, and

entertainment (Shen, Chan, & Hung, 2014). As a relatively new commentary format, Danmaku has become increasingly more popular among Asian countries in recent years including China. It has been widely used in a variety of genres of content, from ACG videos (Anime, Comic, and Game) to live streams, TV series, news, and entertainment shows (Chen, Gao, & Rau, 2015). As shown in Figure 1 (in Appendix A), Danmaku incorporates available user comments and synchronizes them to the video’s specific timestamp, meaning that people who visit the video can also see the comments written by other viewers within the same video image (Johnson, 2013). Once a person posts a comment, it will be immediately available on

the screen, scrolling from right to the left as a part of the content.

Danmaku provides viewers with an opportunity to share instantaneous opinions about the video. By integrating all user comments to the specific time points of the video, this kind of pseudo-synchronic co-viewing system (Johnson, 2013) can enrich the video environment and provide value-added watching experiences to viewers (Lin, Huang, & Cordie, 2018). Danmaku has been seen to provide additional help in understanding the video, with important information related to what is being displayed on a given moment of the video (Lin, et al., 2018; Chen et al., 2015). However, this kind of commentary format may also be perceived as a source of distraction (Chen et al., 2015; Cheng, 2014), because it distracts viewers’ part of attention from original video content to comments. According to Rosenholtz, Li and Nakano (2007), this kind of visual clutter of excessive items being displayed at the same time will lead to comprehension difficulty, degradation of performance, and lower short-term memory. (Alvarez & Cavanagh, 2004).

Despite the increasing use of the Danmaku system in online video sites in many Asian countries, Danmaku is still a new research topic, especially in Western countries. So far, prior studies have agreed on 1) the informational values of Danmaku comments, and 2) the visual clutter problem of Danmaku video. However, few scholars have explored the causality between Danmaku visual presentation and information gain during video viewing. Motivated by this research gap, this study aims to answer the following question:

RQ: How does Danmaku commentary influence a viewer’s information gain from the video content?

Theoretical Background and Hypothesis Development Richness Theory

Media richness theory (MRT, Daft & Lengel, 1986), refers to the media’s ability to deliver communication messages. Daft and Lengel (1986) defined richness as the capacity of

information carrying, the authors argued that richer media not only can provide new understanding but also can clarify ambiguous information more effectively than the media with low richness. Media richness can be measured by four main dimensions including the ability to 1) convey multiple cues to the receiver; 2) facilitate immediate mutual feedback; 3) build a personal focus, for instance, let users get more involved; and 4) use high-variety language cues like emoticon or symbols (Lengel & Daft, 1988). Based on these criteria, media richness theory ranked face-to-face communication as the richest media but the computer-mediated communication to be the leanest media. However, with the introduction of new media, researchers found that social-digital media (i.e., instant messaging software; mobile social television) can also create rich communication context as these applications provide personal, interactive experience to the receivers (Wang, Hsieh, & Song, 2012; Chang & Lee, 2014;).

Media richness is often related to increased information acquisition (Dineen, Ling, Ash, & DelVecchio, 2007). Lengel and Daft (1986) claimed that communication media varies in “the ability of information to change understanding within a time interval” (p. 7), that was defined as information richness. It is argued that rich media let receivers spend less time and effort to comprehend information, especially those information with greater ambiguity and uncertainty. However, too much information cues will degrade the effectiveness of

communication (Otondo, Van Scotter, Allen, & Palvia, 2008). If the amount of information exceeds the threshold that the medium (i.e., video) could process, some information would not be well transmitted to the receivers. Otondo et al. (2008) explained this as information overload. This phenomenon may occur particularly when a media conveying complex information, or the information is delivered at high dense or high speed that the receivers are unable to process.

viewing context because this commentary system presents user comments along with the video content synchronously, enables timely bi-directional feedback, creates higher social presence (Shin, 2013), encourages viewers to use symbolic language, and enhances the perceived interactivity (Liu et al., 2016). The Danmaku video is a type of user-generated media (Chen et al., 2015). Comments that display on the screen along with video images can be regarded as re-creations of original content, however, abundant comments that covering the screen may induce information overload and thus decrease communication effectiveness (Otondo et al., 2008).

Cognitive Overload

Lang’s limited capacity model (LC4MP Model, 2000) asserts that individuals have limited cognitive resources for information processing. In other word, people cannot allocate resources to all perceived information at the same time. Three sub-processes: encoding, storage, and retrieval, are working simultaneously while sharing one single resource pool (Lang, 2006). By encoding, perceived information will be created as mental representation and stored in the brain. Storage refers to the process of linking the current information to an old piece of information that has been previously stored. Retrieval means retrieving this previous information. If too many cognitive resources are allocated to encoding, people will have trouble in storing and retrieving information. When too many messages that a person needs to process within a short period time, the cognitive overload will occur, and the memory performance will then be degraded (Sweller, 1988).

Watching Danmaku videos requires more cognitive resources for information

processing than non-Danmaku videos, as viewers have to process both original video content and the comments message at the same time. Besides that, Danmaku video sites allow users to send comments with different text colors, font size, and also let them send emoticons. When many colorful, animated comments flying across the screen, it adds another layer of

complexity for a video. Danmaku draws additional visual attention from the original video and demand more cognitive resources for encoding, leaving fewer resources for storing and retrieval of information (Lang, 2006). Therefore, too many Danmaku comments on the screen can cause cognitive overload, which will negatively affect the viewer’s memory performance and new information learning.

Drawing on richness theory and the perspective of cognitive overload, this present study compares different levels of Danmaku density (i.e., the number of available Danmaku commentary presented on the screen) and investigates how the visual presentation affect viewers’ information gain from video content.

Danmaku Density and Information Gain

Richness theory (Daft & Lengel, 1986) implies that Danmaku videos are expected to transmit richer information more effectively than non-Danmaku videos. How much

information viewers can acquire from watching Danmaku video is what we called

“information gain”. Danmaku commentary not only helps viewers to understand the complex plot but also enhance their knowledge of certain concepts or issues that viewers might not familiar with (Chen, Gao, & Rau, 2017). Viewers can acquire specific, real-time information about the current video such as hidden meanings of a joke, name of the background music, or the translations of the foreign videos. Therefore, it is reasonable to believe that, with the help of Danmaku commentary, viewers will have an enhanced comprehension of the video content and can learn other relevant information from user comments. Prior studies have tested the informative effect of Danmaku in the educational field. For example, Lin et al., (2018) did an exploratory study and tested the effectiveness of Danmaku for knowledge acquisition during video-based lectures, authors have concluded that Danmaku commentary could help to increase lecturer-to-student interaction and to let students obtain additional knowledge when having video lectures. Leng, Zhu, Wang & Gu (2016) compared student learning performance

in two different conditions (watching Danmaku videos and normal videos). The results showed a significant improvement in the experiment group that watched Danmaku videos, because of Danmaku’s additional explanations and support for video materials.

Compared to the conventional commentary system that viewers need to read all the comments page by page, Danmaku evolves full media capabilities of high transmission velocity (a densified display and rapid delivery of comments), high parallelism (simultaneous viewing experience), and the usage of symbol sets (Liu et al., 2016). However, this can also cause a potential for information overload. Augmented information may interfere with or even harm the original content, and therefore affect the communication outcome

(Chorianopoulos & Lekakos, 2008). Based on Lang’s limited capacity model (2000), an excess of comments overlaid on the screen will add cognitive load on viewers and lead to lower memory performance and lower levels of knowledge enhancement. A previous study by Ni (2017) has tested the effect of Danmaku on cognitive overload, Ni compared three conditions of text transparency (100%, 50%, and 0%) and found a higher memory accuracy in the 50% transparency condition compared to the 100% transparency condition. This might because it was easier for viewers to encounter cognitive overload at the higher

text-transparency condition. In another survey study from Yao, Bort, and Huang (2017), 85% of participants reported visual clutter as the main drawback of Danmaku. They felt annoying to watch a video with large-quantity of comments passing over the screen.

In light of previous studies, this study compared the information gain outcome in different Danmaku visual presentation conditions, which are, high density and low density of Danmaku presentation. High density of Danmaku visual presentation indicates high quantity of comments and therefore a crowded display on the screen, while lower level of Danmaku density indicates less number of comments and a less crowded display on the screen. The effect of Danmaku on information gain is mainly shown as two aspects in the current study:

memory performance and knowledge enhancement. Therefore, the first hypothesis has been proposed:

H1: Higher levels of Danmaku density will decrease information gain from video content. More specifically:

H1a: Higher levels of Danmaku density will decrease memory recall performance. H1b: Higher levels of Danmaku density will decrease knowledge enhancement. Danmaku Density and Distraction

Shifting of the attention from an area of focus to another places or stimuli is a process of distraction. According to Schumm and Post (1997), distraction can be caused by both internal factors (i.e., wandering thoughts or physical illness) and external stimuli (i.e., the greater intensity or attractiveness of another object). Liang and Lee (2010) defined distraction as a conflict between the focused task and the secondary task in their study about driving distraction. The authors explained that both visual distraction (i.e., when the driver looks away from the road) and cognitive distraction (i.e., when the driver interrupts his focus on the road and processes other information) would undermine a driver’s performance.

Danmaku commentary makes video-viewing to be a social experience where

individuals can communicate within a community of viewers (Liu et al., 2016). Prior studies showed that people are more likely to get distracted and cannot be focused enough when multiple partners are communicating simultaneously (Schatz, Wagner, Egger, & Jordan, 2007). Thus, Danmaku comments can be seen as an external source of distraction. In

Danmaku-video-viewing, both visual and cognitive attention of viewers could be drawn away by this kind of “flying” comments. Danmaku asks for the constant back-and-forth eye gaze between the comments and the video content, so viewers cannot be fully focused on the original video. Besides, it also requires viewers to allocate a part of cognitive resources to process the Danmaku comments while watching the videos. Explained by richness theory

(Daft & Lengel, 1986), richer media may contain larger amount sources of distraction, which will distract viewers’ ability to accept and process important information (Badger, Kaminsky, & Behrend, 2014). When encounter too many comments above the screen, viewers’ focus will become more easily to be shifted and may pay more attention to the comments rather than to the video content (Liang & Lee, 2010). Therefore, viewers may feel more distracted when watching a video with larger amount of Danmaku on the screen:

H2: Higher levels of Danmaku density will increase the level of distraction. By interrupting the focus from the desired area, distraction often leads to poor task performance. Rosenholtz et al. (2007) claimed that visual clutter can reduce both “visual search performance” (p. 3) and visual information comprehension. Weissman, Warner, and Woldorff (2004) did a study in classrooms and found that increased levels of distraction have led to decreased classroom participation and lower test grades. A study (Matarazzo & Sellen, 2000) that focused on technology use at the workplace has found a “distraction effect” on job performance. The results demonstrated that remote employees felt distracted when using a new video-conferencing system, which decreased the satisfaction of communication and let them spend longer time to finish the tasks. Alboher (June 2008) supported above studies and provided the evidence that, on average, workers have taken around a half an hour to continue their previous tasks once they get distracted. This is negatively related employees’ working performances. As explained by limited capacity model (Lang, 2000), viewers are not able to allocate all of the available resources to their tasks because of the limited capacities in

information processing (Lang, 2006). That is, too many Danmaku comments above the screen may distract part of viewers’ attention from the original video content and require them to process two sources of information simultaneously. This will add cognitive load on viewers, and in turn interfere with information gain (Shen et al., 2014; Chen et al., 2017). Hence, this hypothesis proposed:

H3: The relationship between Danmaku density and information gain is mediated by the levels of distraction. More specifically:

H3a: The relationship between Danmaku density and memory recall is mediated by the levels of distraction.

H3b: The relationship between Danmaku density and knowledge enhancement is mediated by the levels of distraction.

The Role of Danmaku Experience

An individual’s Danmaku experience may also affect viewers’ watching experience. To understand the hindrance of using Danmaku, Chen et al. (2015) conducted two focus group studies with a total of 11 participants. They found that higher levels of Danmaku density affected experienced viewers less, compared to the users who have less or no

experience in watching Danmaku videos. Experienced users thought it is interesting to watch a video with “a thick covering” of user comments on the screen, while non-users complained that too abundant information would distract them from video viewing, and destroy the aesthetics of a video. While the sample size of this study was too small, it provided the initial evidence that distraction caused by information abundance is the main hindrance for non-users to watch Danmaku videos. In 2017, the authors expanded this initial study and

conducted a survey with 248 participants. The results confirmed again that frequent viewers of Danmaku videos did not feel the visual clutter problems while infrequent viewers treated abundant comments as a distraction source and an additional cognitive load. Yet, another study from Ni (2017) tested the moderator role of Danmaku experience when understanding the effect of Danmaku comments on viewers’ watching experiences, but no significant effect was found among three text-transparency conditions (0% vs. 50% vs. 100%).

With the mixed research results, the current study strives to examine the role of previous experience with Danmaku videos when understanding the relationship between

Danmaku visual presentation and distraction outcome. Distraction occurs when less attention paid to the video content. Having large amounts of comments scrolling across the screen, the original video content that behind the moving comments will become invisible or partly invisible (He & Tang, 2017). Compared to the experienced viewers, people who are rarely watching Danmaku videos may feel more difficult to balance the attention between two sources of information and may get distracted more easily. Therefore, it can be assumed that the frequency of watching Danmaku videos can moderate the Danmaku clutter effect on the distraction. So, the final hypothesis would be:

H4: The relationship between Danmaku density and distraction is moderated by the viewers’ previous experiences of watching Danmaku videos. The effect of Danmaku density on distraction will be stronger for people who have fewer experiences watching Danmaku videos than experienced users.

The conceptual model is illustrated in Figure 2.

Figure 2. Conceptual model

Methodology

In order to answer the research question and to test the proposed hypotheses, the present study will conduct an uni-factorial between-subjects experiment design. Two conditions would be higher density vs. lower density Danmaku.

Participants

This study applied a convenient sampling approach, young native Mandarin speakers (aged from 18 to 30) were invited. Two main reasons can explain why this study has focused on this sample group. Firstly, Danmaku has been increasingly gaining popularity in China in recent years, with many wide-reaching online video sites (i.e. tencent.com) introducing the Danmaku feature to online viewing. Therefore, this study decided to focus on Chinese market and only recruited Mandarin-speaking viewers. Secondly, according to Bilibili Users Annual Report 2017 issued by JiGuang Data (September 2017), a leading mobile data solution platform in China, young adults are heavy users of Danmaku video. 92.8% of users are under 30 years old, of which users aged 20-24 accounts for 53.6%. So, this study decided to target young adults, aging from 18-30 years, as research sample. A total of 155 participants took part in the experiment. All participants were within the target age range. However, 30 participants have dropped out and 13 participants did not watch the video at all. Eventually, this study consisted of 112 participants (68 Female and 44 Males), the average age was 26.29 years old (SD = 2.65). Almost 95.5% of the participants own a bachelor or higher degree. (See Table 1)

Procedure

An online experiment was conducted in the current study, which was approved by University of Amsterdam Communication Faculty. The present study was using convenient sampling for data collection, from November 16, 2018, to November 23, 2018. A survey link from Qualtrics (Qualtrics.com, an online data collection and analysis site) was distributed in the network of the researcher or posted on Chinese social media platform WeChat. Each participant was compensated for around 1 Euro (8 in Chinese currency) for participation via WeChat Pay, which is a popular mobile payment application embedded in WeChat and the received money can be used for any kinds of online or offline consumption in China.

The whole survey took around 10 minutes to complete. The selected video was presented either with high-density Danmaku or low-density Danmaku commentary,

participants were randomly assigned to one of the two conditions. Starting with a short study instruction, participants were told that this study is going to understand viewers watching experience of Danmaku video, participants have to be between 18-30 years old. After the briefing, participants were invited to watch a short clip of video. Viewing times that are too short for watching the video completely were regarded as invalid. Afterward, the dependent variables, control variable, and relevant demographics were measured. Finally, after a manipulation check, the survey ended with a short debriefing about the goal of the study. Experimental Stimulus

This study used a 3-min, 37-sec short clip from a Chinese TV series named First Half of My Life that released in 2017. The choice of series was based on topic and popularity: firstly, this is a series about career, love and how people get through life struggling in big cities, which can be related to by most of the young Chinese people. Secondly, it is not the “most-viewed” drama to avoid possible influence of previous knowledge with the series. Since First Half of My Life is available on bilibili.com, which is the largest Chinese Danmaku video site, and it already have many Danmaku comments. It was a convenient choice to adopt the video with existing comments for this experiment. This video clip contains a complete short sequence of a story. A relatively neutral scenario was chosen where two main characters are talking about job selection. The Danmaku density level (high or low) of the video clip was first set on bilibili.com, the two versions of the stimulus were then recorded by QuickPlayer software.

Measurements

Independent variable. Danmaku density is the independent variable of this study. Participants were watching the video either with high screen density or with low screen

density. Given a total of 381 comments, the video with 70% of the available comments was defined as high-density condition; while the video with 30% comments on the screen was defined as low-density condition. In some certain scenes, participants who exposed to high-density video can see much more comments on the screen. For example, in the same video image at 2m 46s, 21 comments are appearing on the screen in high-density condition, but only three comments are presented in the low-density condition (see Figure 3 & 4 in Appendix A). Dependent variables. Information gain was measured by two different sub-variables, which are memory recall and knowledge enhancement.

To measure the accuracy of memory recall across two conditions, three specific content-related questions about the videos (i.e., not the Danmaku commentary) were asked. Participants had to choose only one single answer per question among the offered three answer options. Answers were first recoded as 1 (True) and 0 (False), then an overall

percentage of memory accuracy for each participant was calculated by averaging all memory test items (M = 84.23%. SD = 23.23%).

Knowledge enhancement was measured by four items and assessed with a seven-point Likert scale ranging from “1 = Strongly disagree to 7 = Strongly agree”. First three items were evaluated by asking participants to give their opinions towards the statements “Danmaku gives me more information about the relationship development between main characters”, “Danmaku comments helped me resolve some content related questions” that adapted from previous study (Yao et al., 2017) and “I have learned new information by watching this Danmaku video”. One additional item that based on technology acceptance model (Park, 2009) was measuring Danmaku usefulness in particular (“Danmaku comments helped me improve my understanding of the story depicted in the video”). These four items were put into Principal Axis Factoring analysis and result turned out to be one single factor. Only one component’s eigenvalue above 1 (eigenvalue 2.89) and a clear point of inflexion

can be seen after this component in the scree plot. A scale was created by having the mean score of the four items, M = 3.67. SD = 1.00, where higher value refers to higher levels of knowledge enhancement. The reliability test showed that this scale was reliable (Cronbach’s alpha = .87).

Mediator. Distraction is the assumed mediator in this study and assessed on seven-point scale (1 = Not at all, 7 = Very much). The first two questions were measuring

participants’ concentration level, items like “How much did you pay attention to the video content you just watched?” and “How much did you concentrate on the video content you just watched?” were adapted from the previous study (Potter & Choi, 2006). Moreover, one question was added by directly asking participants “How distracted were you from the content of the video because of the Danmaku comments?”. The first two concentration items were reverse coded so that lower scores refer to lower levels of distraction while higher scores indicate higher levels of distraction. The result of Principal Axis Factoring analysis of these three items generated one single dimension, only one component’s eigenvalue above 1 (eigenvalue 1.96) and a clear point of inflexion can be seen after this component in the scree plot. The reliability of this scale was reasonably reliable, Cronbach’s alpha = .73. A scale was created by averaging these three items, M = 3.71, SD = .87.

Moderator. Participants’ previous experience of watching Danmaku video (M = 3.27, SD = 2.04) was measured by asking a question “How frequently do you watch Danmaku videos?”, which rated with seven-point scale ranging between “1 = Never” to “7 = Always”.

Control variable. Familiarity with the series (M = 3.83, SD = 2.25) is considered to be a control variable in this study. Familiarity was measured by a question “How familiar are you with this TV series?”, rated by seven-point scale (1 = Not at all familiar, 7 = Very much familiar).

density of Danmaku comments over two experiment conditions, a question item “How busy did you find the level of Danmaku in the video you just watched?” was asked with a seven-point assessment scale (1 = Very much busy, 7 = Not at all busy).

Results of reliability tests for measurement scales were summarized in Table 2 in Appendix B.

Results Manipulation Check

In the current study, participants were randomly assigned to watch either high-density or low-density Danmaku video. Eventually, 56 participants completed each condition.

Independent t-test was conducted for manipulation check analysis with Danmaku density (High vs. Low) as the independent variable and perceived density level as the dependent variable. People who were assigned to high-density condition (M = 5.39; SD = 1.33) indicated significantly higher Danmaku busy level on the video screen than people who were asked to watch low-density Danmaku video (M = 3.41; SD =1.64). The mean difference between two groups is 1.98, t (105.57) = 7.03, p < .0001, 95% CI [ 1.42, 2.54]. The manipulation was successful in the current study, participants have correctly recognized the density level of Danmaku comments on the video screen.

Randomization Check

To check whether age was comparable over the two Danmaku conditions, an

independent t-test was conducted with Danmaku density (High vs. Low) as the independent variable and age as the dependent variable. The results showed that mean age of participants in the low-density Danmaku condition (M = 26.09, SD = 2.88) was not significantly different from the participants’ mean age in the high-density condition (M = 26.48, SD = 2.39),

Mdifference = -.39, t (106.48) = -.78, p = .435, 95% CI [ -1.39, .60]. Therefore, age was equally distributed in two experimental groups.

experimental conditions. As a result, it appeared that both gender (χ2 (1, N = 112) = 2.40, p = .122) and education (χ2 (3, N = 112) = 3.46, p = .326) were equally distributed over two conditions.

Hypothesis Testing

Before the analysis, the experimental condition with low-density Danmaku was coded as 0 while the high-density Danmaku condition was coded as 1. The SPSS macro PROCESS by Hayes (version 3.0) model 7, with 5000 bootstrap samples, has been ran twice with memory recall and knowledge enhancement as two separate dependent variables. In this moderated mediation analysis, Danmaku density was served as the independent variable, previous experience with Danmaku video as the moderator, and the level of distraction as the mediator. The control variable, familiarity with the series, was kept controlled during the analysis. Table 3 (in Appendix B) gives descriptive statistics of variables, while Table 4 (in Appendix B) shows the results from PROCESS model 7. Results of the tested model was depicted in Figure 5.

H1 proposed a negative effect of Danmaku density on viewer’s information gain. Firstly, for memory recall performance, when controlling for other variables, Danmaku density showed a marginally significant effect on memory recall, b = -.10, t = -2.27, p = .025, 95% CI [ -.20, -.01]. Viewers in high-density Danmaku condition showed .10 lower memory recall performance compared to viewers who exposed in low density condition. H1a was supported. Secondly, controlled for other variables, Danmaku density showed a significant effect on knowledge enhancement, b = -.44, t = -2.20, p = .03, 95% CI [ -.84, -.04]. Viewers in high-density Danmaku condition reported .44 lower levels of knowledge enhancement compared to viewers in low density condition. H1b was supported. In conclusion, Danmaku visual density significantly decreased individual’s memory recall performance and the level of knowledge enhancement. H1 was thus supported. Higher density of Danmaku commentary

has decreased viewer’s information gain compare to the lower density of Danmaku commentary.

H2 assumed that higher levels of Danmaku density will increase distraction. The result from Model 7 confirmed a significantly effect of Danmaku density on distraction, b = .69, t = 2.32, p = .022, 95% CI [ .10, 1.29]. Viewers in high-density Danmaku condition indicated .69 higher levels of distraction compared to viewers in low density condition. Therefore, H2 was supported. H3 predicted a mediation effect of distraction in the relationship between

Danmaku density and information gain. However, the mediator (distraction) did not

significantly affect memory recall and knowledge enhancement. For memory recall, b = -.04, t = -1.45, p = .151, 95% CI [ -.09, .01]. For knowledge enhancement, b = .00, t = .03, p = .974, 95% CI [ -.22, .23]. Although, the path from Danmaku density to distraction and the path from Danmaku density to information gain were both approved, the non-significant effect of mediator (distraction) on dependent variables suggesting an invalid mediation effect. The indirect effect of Danmaku density on memory recall through distraction levels was not significant, b = -.02, 95% CI [-.06, .00]. The indirect effect of Danmaku density on

knowledge enhancement through distraction levels was not significant either, b = .00, 95% CI [-.14, .14]. H3a and H3b were both not confirmed. Therefore, H3 was rejected. Danmaku density did not indirectly affect individuals’ information gain through the mediating cause of the perceived distraction.

H4 assumed that the relationship between Danmaku density and distraction is

moderated by the viewers’ Danmaku experiences. However, the moderation role of Danmaku experiences was not found. When controlling for other variables, viewers’ previous Danmaku experiences did not significantly affect the level of distraction, b = .03, t = .53, p = .596, 95% CI [ -.08, .14]. An interaction effect between Danmaku density and previous experience was not found either, b = -.02, t = -.29, p = .771, 95% CI [ -.18, .13]. Therefore, H4 was rejected.

Previous experience with Danmaku video did not moderate the relationship between Danmaku density and the levels of distraction.

Figure 5. PROCESS Model 7 result

Discussion

This study investigated whether Danmaku visual density has different impacts on individuals’ information gain during video-viewing, by comparing two Danmaku presentation conditions (with low density vs. high density). The role of perceived distraction and previous experiences with Danmaku videos have also been explored to understand Danmaku-video-viewing. The present study assumed that Danmaku density will negatively affect individuals’ information gain and will positively affect the perceived distraction. It is also proposed that the level of distraction will mediate the relationship between Danmaku density and

information gain. Individuals’ previous experience was introduced as a moderator, assuming it will moderate the effect of Danmaku density on distraction. However, results only

supported the direct effect of Danmaku density on information gain and on distraction. Both mediation and moderation effect were not found in this study.

Results showed that people who were exposed to the high-density Danmaku condition recalled less information after video-viewing than people who were exposed to the low-density Danmaku condition. It resonates with the findings of previous studies (Potter, 2000) that individuals are less able to store and retrieve information when faced with overwhelming messages through an online interactive medium, because of individuals’ limited cognitive resources (Lang, 2006) and limited working memory (Mayer, 2009). Results also revealed that higher density of Danmaku presentation led to a lower level of knowledge enhancement. Explained by previous studies (Chen et al., 2015; Chen et al., 2017), Danmaku commentary enhances viewers’ understanding and knowledge of the video by supplying plot explanations or alternative video-related information. However, dealing with abundant information

simultaneously may cost more cognitive resources of viewers to evaluate the received messages, which in turn would increase the comprehension difficulty of the video and influences viewers’ information acquisition during video-viewing (Badger et al., 2014).

This study also found that participants in high-density condition reported a higher level of distraction than those who were exposed to low-density condition. Since viewers can only allocate their attentions to a small part of the incoming information at once (Baddeley, 1992), the appearance of Danmaku comments can be interpreted as a distraction stimulus and a source of annoyance. Consequently, large amount of comments presented on the video screen would direct viewers’ more attention away from video content. However, contrary to the limited capacity model (Lang, 2000) which argues that distraction caused by cognitive overload impairs one’s memory performance and information learning, distraction did not lead to low information gain in this study. This might be due to the fact that the measurement of distraction in the current study did not measure the real cognitive load. Multiple resources theory (Wickens, 2002) defined distraction to be different types such as verbal, visual or cognitive distraction etc., among which, cognitive distraction is more related to cognitive

overload. Although the result of this study found that higher density Danmaku was perceived as more distracting, it is hard to say that whether participants were more cognitively distracted or more visually distracted. The content of Danmaku might be another factor to explain these results. Suggested by previous study (Zhang, Qian, Pi, & Yang, 2018) that participants who viewed Danmaku that not related to the content reported lower learning achievement in online video-learning compared to those who viewed the content-related Danmaku. In this study, high-density video contains more humorous comments which are irrelevant to the content (i.e. “hahaha” or “2333”) but less content-related informative comments such as helpful

explanations for the video story. This might be a potential influence of the lower levels of both memory recall and knowledge enhancement in high-density condition.

As for the experiences with Danmaku video, previous qualitative studies (Chen et al., 2015; Chen et al., 2017) concluded that the infrequent viewers felt highly distracted by abundant comments while the experienced users did not feel bothered at all. However, the current study was not able to confirm such results. Compared to the frequent viewers, participants who rarely watch Danmaku videos did not get distractive more in the high-density condition. One explanation would be that sample size of experienced viewers was much less than inexperienced viewers (around 70% are infrequently users of Danmaku) in each experimental group, which might potentially influence the result. Future studies should include more experienced viewers of Danmaku video to test the effect. Secondly, Danmaku-supported video sites normally allow users to turn off the Danmaku at any time during video-viewing. In real life, viewers often turn off the Danmaku when they feel they are being

interfered with Danmaku or the video content is not suitable to have Danmaku. So, it does not mean that viewers who are often watching Danmaku video will be more comfortable with visual clutter. It is possible that frequent viewers are as vulnerable as infrequent viewers towards Danmaku at high-density.

Theoretical Implications

This study makes important contributions to literature. Firstly, it has extended the application of richness theory to online video environment. Although, previous scholars considered computer-mediated communication to be the leanest media (Daft & Lengal, 1986), evidence shows that the criteria might be different when assessing the richness of a digital media product in the current society (Anandarajan, Zaman, Dai, & Arinze, 2010; Sheer, 2011). Danmaku commentary system adds social functionality to a regular video site, which makes personal, interactive communication possible. Viewers can send comments while watching video and receive immediate feedback from other co-viewing viewers. Emoticons are also supported in Danmaku videos. Although, it lacks verbal cues or facial expressions, Danmaku video still should be considered as a rich medium as it creates a virtual immersive co-viewing context and provides with audiences the enhanced content.

Secondly, media richness theory (Daft & Lengal, 1986) argues that richer medium performs better at communication effectiveness. However, the increase in information richness may negate the benefits of rich media (Dineen et al., 2007). By comparing different density levels of Danmaku, this study highlighted the negative side of information richness in Danmaku context. Participants who were exposed to the video with low-density condition performed better at memory test and learned more additional information.

Information richness often relates to the problem of overload. The findings also support limited capacity model and the perspective of cognitive overload (Lang, 2006). Because of the scarce cognitive resources, individuals only can devote part of their attention to the received information at one time. This study provides the evidence that viewers in high-density condition were less able to concentrate on original video, and felt hard to comprehend two sources of information. The current study incorporates distraction as a key variable to understand the influence of Danmaku on watching experience. Although, unexpectedly, the

distraction effect on participants’ information gain was not confirmed in the current study. It did reveal that Danmaku comments interfered with the primary task of watching video, participants have become less focused on the original video and felt more easily to be distracted in the face with large amounts of stimulus above the screen.

Practical Implications

This study also has some practical suggestions. Firstly, many Danmaku-supported video sites are currently focusing on the social functionality of Danmaku, while its

informative value should also be emphasized. Integrating valuable comments with video is a good attempt to enhance the video content, to increase user experience, and to improve video views. However, considering the limited carrying capacity of the video itself and the limited cognitive resources of viewers, it is essential to pay attention to the problem of visual clutter. Danmaku video sites in China like bilibili.com have implemented many sorts of

customization options, such as the edit tools of font size, font color, display zone and text transparency, by which users can self-set the style of Danmaku presentation. However, these fancy tools might also increase one’s perceived difficulty of Danmaku usage. It is necessary to build a smarter filter system, such mechanism should be able to control the quantity of Danmaku of each video and automatically block the repeated, irrelevant or low-quality comments.

Secondly, one screen sharing two sources of information is inevitably influencing viewers’ focused attention. Designers of online video sites should optimize the user experience by reducing the potential distracting effect. For example, leave the most useful comments on the screen or only let comments appear on the upper side of the screen. Moreover, setting up a reporting system and encouraging viewers to report low-quality messages are also helpful to reduce the disturbing experience. However, this study did not find the negative effect of distraction on video-related information acquisition. It might not be

necessary to exaggerate the dark side of Danmaku. Video-relevant comments will add values to both the video content and viewers (Zhang et al., 2018).

Thirdly and finally, the findings also have some implications on the offline application of Danmaku. Except for online videos, many cinemas have tried to use Danmaku as a selling point to attract more audiences and increase the box office in recent years. However, since Danmaku will capture a part of viewers’ attention from the film, cinemas should balance the benefits and the risk. For example, instead of applying Danmaku to all projection rooms, opening one special room for viewers who love watching Danmaku might be a solution. Limitations and Future Studies

Some limitations of this study need to be noticed. First of all, this study used convenient sampling to recruit participants. Although the non-random sampling method is time-saving, the sample might not be representative enough. Therefore, the population validity would be lower in this study. Secondly, this study focused on young adults between 18-30, but 75% of the participants are over 25 years old. The unequal distribution of age might influence the result. 95s Online Behavior Insights Report (Tencent Kueclub, June 2018) revealed a youthful tendency of Danmaku viewers, young people aged between 18-24 years old are more interested in Danmaku. So, future research can target a younger group of viewers to test the effect. Thirdly, this online experiment manipulated the Danmaku

presentation density over two conditions. Participants were “forced” to watch the video and were not able to change anything about its presentation, which did not reflect the real situation of Danmaku viewing and thus lower the external validity. Moreover, the use of self-report approach may cause various biases as people often overestimate or underestimate their performances. For example, the self-reported distraction may not actually measure the real distraction level. In a combination of the subjective information, future studies could also include objective tests like eye-tracking of attention allocation for getting more reliable data.

Additionally, the content of Danmaku has not been taken into account in this study. As mentioned before, different types of Danmaku comments, for instance, informative

content vs. entertainment content, may have different impact on one’s information acquisition and watching experience. Future studies should consider the influence of comment content. Besides, the differences of comments content and presentation style (i.e., font size, font color, etc.) between two conditions should be controlled to exclude any other potential influence. Finally, although most of the measurement items used in this study were adapted or revised from the existing scale, some questions were self-created. The reliability of the scale should be further tested in future studies.

Conclusion

The implementation of Danmaku system in online video is an increasingly common phenomenon. Danmaku as a user-generated content makes online co-viewing possible. By integrating user comments onto video screen, Danmaku helps in delivering rich information to viewers. It can not only boost viewers’ memory but also enhance their knowledge of the video content. However, the more Danmaku on the screen does not necessarily mean the better. Firstly, it should be noticed that too much information may cause cognitive overload. When fewer recourses could allocate to the video content, viewers’ information gain would be affected. Secondly, visual clutter affects one’s watching experience. Abundant comments on the screen may attract a part of attention and let them not be able to fully focus on the original video content. In summary, to provide a healthy video-viewing environment, the problem of visual clutter needs to be solved. This study suggests that the quantity of screen comments should be controlled within a reasonable range so that the Danmaku can better exert its informative value. Meanwhile, a smarter filter system should be built to block low-quality information and ensure more useful, valuable comments to be contained on the video screen.

References

Alboher, M (June 2008). Fighting a War Against Distraction. Retrieved from: https://www.nytimes.com/2008/06/22/jobs/22shifting.html

Alvarez, G. A., & Cavanagh, P. (2004). The capacity of visual short-term memory is set both by visual information load and by number of objects. Psychological science, 15(2), 106-111.

Anandarajan, M., Zaman, M., Dai, Q., & Arinze, B. (2010). Generation Y adoption of instant messaging: An examination of the impact of social usefulness and media richness on use richness. IEEE Transactions on Professional Communication, 53(2), 132-143. Baddeley, A. (1992). Working memory. Science, 255(5044), 556-559.

Badger, J. M., Kaminsky, E. S., & Behrend, S. T., (2014). Media richness and information acquisition in internet recruitment. Journal of Managerial Psychology, 29(7), 866-883. Chang, Y. L., Lee, C. M., (2014). The Impacts of LINE Sticker Values, Fashion Involvement,

and Network Externality on Sense of Virtual Community and Stickiness – the Viewpoints of Media Richness and Social Presence Theory (CN). Electronic

Commerce Studies, 12(4), 419-449.

Chen, Y., Gao, Q., & Rau, P. L. P. (2015, August). Understanding gratifications of watching danmaku videos–videos with overlaid comments. In International Conference on

Cross-Cultural Design (pp. 153-163). Springer, Cham.

Chen, Y., Gao, Q., & Rau, P. L. P. (2017). Watching a movie alone yet together: understanding reasons for watching Danmaku videos. International Journal of

Human–Computer Interaction, 33(9), 731-743.

Cheng, G. (2014). Exploring students' learning styles in relation to their acceptance and attitudes towards using Second Life in education: A case study in Hong

China Netcasting Services Association (November 2018). China Network Audiovisual

Development Research Report 2018. Retrieved from: http://www.cnsa.cn/index.php/industry/industry_week.html

Chorianopoulos, K., & Lekakos, G. (2008). Introduction to social TV: Enhancing the shared experience with interactive TV. Intl. Journal of Human–Computer Interaction, 24(2), 113-120.

Daft, R. L., & Lengel, R. H. (1986). Organizational information requirements, media richness and structural design. Management Science, 32(5), 554-571.

Dineen, B.R., Ling, J., Ash, S.R. & DelVecchio, D. (2007). Aesthetic properties and message customization: navigating the dark side of web recruitment. Journal of Applied Psychology, Vol. 92 No. 4, No. 2, pp. 356-372.

He, Y., & Tang, T. Y. (2017). Recommending highlights in Anime movies: Mining the real-time user comments “DanMaKu”. Intelligent Systems Conference (IntelliSys),

2017 (pp. 319-322). IEEE.

JiGuang Data (September 2017). Bilibili Users Annual Report 2017. Retrieved online: https://www.jiguang.cn/reports/106

Johnson, D. (2013). Polyphonic/pseudo-synchronic: Animated writing in the comment feed of Nicovideo. Japanese Studies, 33(3), 297-313.

Lang, A. (2006). Using the limited capacity model of motivated mediated message processing to design effective cancer communication messages. Journal of

Communication, 56(suppl_1), S57-S80.

Lang, A. (2000). The limited capacity model of mediated message processing. Journal of

Leng, J., Zhu, J., Wang, X., & Gu, X. (2016). Identifying the Potential of Danmaku Video from Eye Gaze Data. In Advanced Learning Technologies (ICALT), 2016 IEEE 16th

International Conference on (pp. 288-292). IEEE.

Lengel, R. H., & Daft, R. L. (1988). The selection of communication media as an executive skill. Academy of Management Perspectives, 2(3), 225-232.

Liang, Y., & Lee, J. D. (2010). Combining cognitive and visual distraction: Less than the sum of its parts. Accident Analysis & Prevention, 42(3), 881-890.

Lin, X., Huang, M., & Cordie, L. (2018). An exploratory study: using Danmaku in online video-based lectures. Educational Media International, 1-14.

Matarazzo, G., & Sellen, A. (2000). The value of video in work at a distance: Addition or distraction? Behaviour & Information Technology, 19(5), 339-348.

Mayer, R. E. (2009). Multimedia Learning. Cambridge University Press.

Ni, Y. Q. (2017). A Study of Danmaku Video on Attention Allocation, Social Presence,

Transportation to Narrative, Cognitive Workload and Enjoyment. (Master Thesis).

Retrieved from: http://surface.syr.edu/thesis/127

Otondo, R. F., Van Scotter, J. R., Allen, D. G., & Palvia, P. (2008). The complexity of richness: Media, message, and communication outcomes. Information &

Management, 45(1), 21-30.

Park, S. Y. (2009). An Analysis of the Technology Acceptance Model in Understanding University Students' Behavioral Intention to Use e-Learning. Educational Technology

& Society, 12 (3), 150–162.

Pew Research Center (September 3, 2017). About 6 in 10 young adults in U.S. primarily use

online streaming to watch TV. Retrieved from: https://pewrsr.ch/2wZNFC7

Potter, R. F. (2000). The effects of voice changes on orienting and immediate cognitive overload in radio listeners. Media Psychology, 2(2), 147-177.

Potter, R. F., & Choi, J. (2006). The effects of auditory structural complexity on attitudes, attention, arousal, and memory. Media Psychology, 8(4), 395-419.

Rosenholtz, R., Li, Y., & Nakano, L. (2007). Measuring visual clutter. Journal of vision, 7(2), 17-17.

Schatz, R., Wagner, S., Egger, S., & Jordan, N. (2007). Mobile TV becomes

social-integrating content with communications. In Information Technology Interfaces, 2007.

ITI 2007. 29th International Conference on (pp. 263-270). IEEE.

Schumm, J. S., & Post, S. (1997). Executive Learning: Successful Strategies for College

Reading and Studying. Upper Saddle River, NJ: Prentice Hall.

Sheer, V. C. (2011). Teenagers' use of MSN features, discussion topics, and online friendship development: The impact of media richness and communication control.

Communication Quarterly, 59(1), 82-103.

Shen, Y. (2017). Two studies on strategically designed intervention (Doctoral

dissertation).Retrieved from http://scholarbank.nus.edu.sg/handle/10635/136054 Shen, Y., Chan, H. C., & Hung, I. W. (2014). Let the comments fly: The effects of flying

commentary presentation on consumer judgment. International Journal of Systems

Science, 12(12), 1469–1475.

Shin, D. H. (2013). Defining sociability and social presence in Social TV. Computers in

Human Behavior, 29(3), 939-947.

Tencent Kueclub. (June 2018). 95s Online Behavior Insights Report 2018. Retrieved from: https://users.qq.com

Wang, W., Hsieh, J. P. A., & Song, B. (2012). Understanding user satisfaction with instant messaging: An empirical survey study. International Journal of Human-Computer

Weissman, D. H., Warner, L. M., & Woldorff, M. G. (2004). The neural mechanisms for minimizing cross-modal distraction. Journal of Neuroscience, 24(48), 10941-10949. Wickens, C.D. (2002). Multiple resources and performance prediction. Theoretical Issues in

Ergonomics Science 3 (2), 159–177.

Yao, Y., Bort, J., & Huang, Y. (2017). Understanding Danmaku's Potential in Online Video Learning. In Proceedings of the 2017 CHI Conference Extended Abstracts on Human

Factors in Computing Systems (pp. 3034-3040). ACM

Zhang, Y., Qian, A., Pi, Z., & Yang, J. (2018). Danmaku Related to Video Content Facilitates Learning. Journal of Educational Technology Systems, 0047239518811933.

Appendix A – List of Figures

Figure 1. An example of Danmaku Video

Figure 3. Screenshot of Low Density Condition (30%)

Appendix B – List of Tables Table 1

Sample Descriptive Statistics

High Density Low Density Total

Age Age mean 26.48 26.09 26.29 Age SD 2.40 2.88 2.65 Gender Male 28 (46.4%) 26 (32.1%) 44 (39.3%) Female 38 (67.9%) 30 (53.6%) 68 (69.7%) Education Secondary education or lower 3 (5.4%) 2 (3.6%) 5 (4.5%) Bachelor’s degree or equivalent 39 (69.6%) 33(58.9%) 72 (64.3%) Master’s degree or equivalent 14 (25%) 19 (33.9%) 33 (29.5%) Doctoral degree or higher 0 (0%) 2 (3.6%) 2 (1.8%) Total 56 56 112 Table 2 Scale Reliability

Cronbach's Alpha N of items

Knowledge Enhancement .87 4

THE EFFECT OF DANMAKU COMMENTARY 33

Table 4

Moderated mediation result: PROCESS Model 7

Mediation analysis Moderation analysis Memory recall as DV Knowledge enhancement as DV Distraction

b (SE) 95% CI b (SE) 95% CI b (SE) 95% CI Danmaku density -.10* (.05) [-.20, -.01] -.44*(.20) [-.84, -.04] .69*(.30) [.10, 1.29] Distraction -.04 (.03) [-.09, .01] .00 (.11) [-.22, .23] Danmaku experience .03 (.05) [-.08, .14] Interaction -.02 (.08) [-.18, .13] Control variable .00 (.01) [-.02, .02] -.06 (.04) [-.15, .02] -.06 (.04) [-.13, .01] Note: *p<.05; CI = Confidence Interval; values represent unstandardized coefficients

Table 3

Descriptive Statistics of variables

High-density condition Low-density condition Memory recall accuracy

percentage M 77.98% 90.48% SD _{26.42% 17.66%} Knowledge enhancement M 3.42 3.92 SD _{1.02 .93} Distraction M 3.99 3.42 SD _{.78 .87}

Appendix C – Survey Questions

Level of distraction (Developed from Potter & Choi, 2006) * Seven-point scale (1 - Not at all, 7 - Very much):

How much did you pay attention to the video content you just watched? How much did you concentrate on the video content you just watched?

How distracted were you from the content of the video because of the Danmaku comments？ Knowledge Enhancement

* Seven-point scale (1 - Strongly disagree, 7 - Strongly agree):

Danmaku gives me more information about the relationship development between main characters.

Danmaku comments helped me resolve some content related questions. (Adapted from Yao et al., 2017)

I have learned new information by watching this Danmaku video.

Danmaku comments helped me improve my understanding of the story depicted in the video. (Adapted and revised from Park, 2009)

Memory recall test *Single choice

What kind of snack Zijun gave to Hehan in the video?

 Baobaoguo

 Chocolate

 Milk

How many job options Hehan has offered to Zijun in the video?

 One

 Two

Which job option was initially proposed by Zijun’s ex-husband?

 Henglong Square Shopping Mall

 Xintiandi Shopping Mall

 Consulting Company

Previous experience with Danmaku * Seven-point scale (1 - Never, 7 -Always): How frequently do you watch Danmaku videos? Familiarity with the series

* Seven-point scale (1- Never, 7 - Always): How familiar are you with this tv series? Manipulation check

* Seven-point scale (1 - Not at all busy, 7 - Very much busy):

How busy did you find the level of Danmaku in the video you just watched? Demographics

What is your age? ___ What is your gender?

 Male

 Female

What is the highest level of education that you completed？

 Secondary education or lower

 Bachelor’s degree or equivalent

 Master’s degree or equivalent

 Doctoral degree or higher