Introduction

The past decade has witnessed a surge in the popularity of videos for learning, particularly in massive open online courses (MOOCs). However, one of the significant challenges in video-based learning is the passive consumption of information, which results in a passive learning process (Chi & Wylie, 2014). To tackle this issue, interactive computer-mediated communication (CMC) tools are widely used in MOOCs. In such circumstances, it is worthwhile investigating how learners actively utilize these tools during their interactions and what affordances these tools offer to facilitate collaborative learning. This study focuses on two specific tools employed in a MOOC setting: the discussion board and danmaku. Discussion boards provide a forum-like design for discussions. Danmaku, in contrast, allows learners to send messages synchronized with the video timeline while watching.

The rapid growth of scaled learning presents new challenges for researchers within computer-supported collaborative learning (CSCL). First, some CSCL scholars contend that interaction in such nascent contexts fails to meet the strict standards of collaboration widely accepted in the research community (Wise & Schwarz, 2017), whereas other researchers hold a different perspective, arguing for new frameworks to better conceptualize the collaboration in such a larger community (e.g., Jeong et al., 2017). In addition, CSCL researchers are faced with the task of addressing large-scale participation. Typically, this is achieved either by sustaining and harnessing the large scale (e.g., Rosé & Ferschke, 2016) or by reducing it to a smaller-scale scenario (e.g., Wen et al., 2017). Another challenge concerns the lack of socio-emotional presence. The socio-emotional interaction plays a vital role in shaping the development of the learning group and establishing a social space that accommodates the cognitive interaction. However, traditional CSCL tools and environments typically focus on cognitive interaction, neglecting socio-emotional interaction (Kreijns et al., 2003; Kreijns et al., 2013).

This study adopts the Community of Inquiry (CoI) framework developed by Garrison et al. (2000) to understand the cognitive and socio-emotional interaction in danmaku and on discussion boards. To date, the existing literature has extensively explored social interaction and strategies for addressing scalability issues in CSCL. However, there remains a noticeable gap in the research regarding how cognitive and socio-emotional dimensions of social interaction are accommodated within different approaches to handling scalability when utilizing distinct interactive tools, all within the confines of a naturalistic, non-controllable online CSCL environment. Recognizing the significance of nurturing both dimensions of social learning and the imperative to address collaboration within scaled contexts, this study examines the processes underlying scaled collaboration in danmaku and on discussion boards, with a particular focus on the resultant interactions among learners.

The following research questions guide this study:

  • RQ1: How is mass participation accommodated in danmaku and on the discussion board, respectively?

  • RQ2: What content categories of interaction are implemented in terms of socio-emotional and cognitive interactions in danmaku and on the discussion board, respectively? How does learners’ participation differ between danmaku and the discussion board across these categories?

  • RQ3: What are the outstanding interactional behaviors in these categories in danmaku and on discussion boards? How are these behaviors implemented?

Prior research

MOOCs and CoI

Since their emergence in 2008, MOOCs have progressively become one of the foremost trends in open online education (Zhu et al., 2020b). MOOCs provide accessible education to learners with minimal or no financial burden (Siemens, 2013). Nonetheless, despite the considerable learning opportunities offered by MOOCs, they also present various challenges and issues, such as lack of social interaction, low teaching engagement (Siemens, 2013), and difficult learning collaboration (Smith et al., 2011). Conducting collaborative activities can enhance interactivity between learners within the course (Cherney et al. 2018; Rosé et al. 2015), thus potentially mitigating some problems that come with MOOCs. For CSCL, in the face of emerging large-scale learning environments, MOOCs are also attracting increasing attention as a research topic (e.g., Rosé & Järvelä, 2023).

Universities have played the most significant roles in the initiation and development of MOOCs and provide thousands of courses worldwide. In addition, most popular MOOC platforms, such as Coursera and edX, collaborate with universities and provide undergraduate or post-graduate courses for adult learners. However, there are also many MOOCs for K–12 learners, and one example is the high-school-level “Computer Science 101” course analyzed by Kizilcec et al. (2013). Our study uses an open course about high-school-level math knowledge. It is an informal MOOC wherein there is no need to register for the course (but learners need to sign up on the MOOC platform to create an account), and anyone can learn and go.

In online learning environments, the interaction among learners plays a pivotal role (Richardson et al., 2017). The CoI framework has emerged as one of the predominant models for comprehending learning through online interactions (Breivik, 2016) and is recognized as a fundamental context for fostering higher-order learning (Garrison et al., 2001). This framework delineates three interconnected elements—cognitive presence, social presence, and teaching presence—that collectively contribute to a meaningful educational experience (Garrison et al., 2000). It posits that effective online learning depends upon the establishment of a learning community (Swan et al., 2009). Within the context of MOOCs, such a community is traditionally sustained within the course’s discussion forum, but in recent years, it can also be established using other emergent tools. Following its proposal in 2000, scholars have developed different expansions to the original CoI framework, such as additional types of presence (e.g., Kozan & Caskurlu, 2018). In this study, we use the original three types of presence as outlined by Garrison et al. (2000).

Cognitive and socio-emotional interactions in collaborative learning

Collaborative learning encompasses both cognitive and socio-emotional interactions (e.g., Bales, 1999; Kreijns et al., 2013). Cognitive interactions are conceptualized as learners’ interactions about the learning content, learning tasks, or learning process (Dillenbourg et al., 1995; Isohätälä et al., 2020; Järvelä et al., 2016a, 2016b; Kreijns et al., 2003; Kreijns et al., 2013), and socio-emotional interactions are referred to as learners’ interacting with and relating to each other in terms of emotion, motivation, relationship, and so forth (Isohätälä et al., 2020; Kreijns et al., 2003; Kreijns et al., 2013). Cognitive interactions play a fundamental role in successful collaborative learning by facilitating knowledge sharing and construction among participants while nurturing shared meaning-making (e.g., Iiskala et al., 2011; Khosa & Volet, 2014; Näykki et al., 2017b). At the same time, socio-emotional interactions are also crucial, as they provide the foundation for collaboration by offering interpersonal and emotional support (Näykki et al., 2017a; Onrubia & Engel, 2012). Socio-emotional interactions have also been found to have a considerable influence on cognitive interactions (Zhu et al., 2020a). The literature shows that positive socio-emotional interactions can facilitate effective and desirable cognitive interactions (Hu et al., 2021; Su et al., 2018; Zhang et al., 2021). For instance, active socio-emotional interactions cause participants to make greater efforts in collaboration and conduct appropriate activities (Järvelä et al., 2021); a positive socio-emotional atmosphere enables collaborators to resolve conflicts, rejoin in teamwork, and rapidly reach agreement (Hu et al., 2021); and socio-emotional interactions are central to establishing a supportive social space with trust, a sense of belonging, and strong interpersonal connections to facilitate knowledge sharing and joint meaning-making (Brandon & Hollingshead, 1999; Rourke & Anderson, 2002).

In the era of digitalized learning environments, the socio-emotional experiences of both students and instructors have garnered increased attention (Huang et al., 2022). However, empirical research on socio-emotional interactions in collaborative learning remains relatively scarce (Huang & Lajoie, 2023). Furthermore, it is noteworthy that CSCL environments tend to prioritize the cognitive dimension of social interaction while overlooking the socio-emotional dimension (Kreijns et al., 2003; Kreijns et al., 2013).

Collaboration at scale

The nascent context of large-scale learning elicits new considerations regarding collaboration in the CSCL community. Two of the key issues are how to conceptualize collaboration at scale and how to deal with it.

In regard to the conceptualization of collaboration, since its inception, the CSCL community has maintained a strict standard for defining what qualifies as collaboration (Suthers, 2006; Wise & Schwarz, 2017). Participants in collaborative learning often get involved in different group activities, including group negotiation (e.g., Crawford et al., 1999), interactional meaning-making (e.g., Stahl et al., 2014), knowledge co-creation (e.g., Kimmerle et al., 2015), and the coordination of goals (e.g., Weinel & Reimann, 2007). These are essential processes of small-group collaboration but may not always be available in the context of mass participation. Regarding this, researchers in CSCL hold different views: some doubt whether it is necessarily appropriate to implement collaborative learning in large-scale contexts wherein the collaboration widely recognized by the community does not comply well (Wise & Schwarz, 2017), while others offer a new framework for conceptualizing collaboration in such new contexts (e.g., Cress et al., 2013; Elliott, 2016). One representative attempt is the Attendance, Coordination, Cooperation, and Collaboration (A3C) framework (Jeong et al., 2017), which distinguishes between different joint-interaction types in large online knowledge communities.

In terms of dealing with large-scale collaboration, two main strategies can be seen: leveraging the advantages of the scale and addressing the challenges by reducing the scale (Chen et al., 2021). On the one hand, scaled participation can be harnessed to facilitate peer support and collaboration on a level that is not feasible in smaller contexts. For example, in MOOCs, a massive number of learners can be utilized to expedite peer reviews and enable quick feedback from others (e.g., Ferguson & Sharples, 2014; Kulkarni et al., 2015). Platforms such as Wikipedia demonstrate how mass participation can foster collaborative content creation by engaging millions of users (Laniado et al., 2011). On the other hand, strategies have been proposed to address the challenges associated with large-scale participation by reducing the scale. This involves creating smaller group contexts—a well-established realm in CSCL either by employing collaboration scripts to divide learners into appropriate groups for coordinated collaboration activities (e.g., Håklev et al., 2017) or by organizing learners into smaller teams on the basis of their transactive interactions during community-wide discussions (e.g., Wen et al., 2017). It is also important to recognize that large-scale participation presents both challenges and opportunities. Collaborative learning can benefit from designs that mitigate challenges while leveraging benefits. For example, Håklev and Slotta (2017) employed scripts in a MOOC for in-service teachers to utilize the diversity of a large number of participants to form various special interest groups (SIG) based on shared interests and to facilitate small-scale intense collaboration within each group.

Danmaku and discussion boards as technologies for learning and social interaction

Various studies have investigated the affordances and usage of different CMC tools (Cheng & Kinshuk, 2020). In CSCL, which depends mainly on asynchronous communication using text (Kwon et al., 2014), one widely used tool is discussion boards, which enable threaded discussions in a forum design (Cheng & Kinshuk, 2020). In collaborative learning, discussion boards have been shown to effectively support learning interactions without geographic or temporal restrictions, showcasing their flexibility (Hew et al., 2010). Furthermore, they allow learners to take their time to reflect on others’ posts and provide thoughtful responses (Pena-Shaff & Nicholls, 2004). However, discussion boards also have shortcomings. For example, a lack of social interaction has been identified in these forums (e.g., Kreijns et al., 2013), and learners’ engagement with the forum can be relatively low (e.g., Corrin et al., 2017).

In recent years, the novel feature of online videos, known as danmaku, has been steadily gaining attention. This communication system originated in Japan and has gained significant popularity in Asia (Zhang & Cassany, 2019). Danmaku is characterized by messages that are synchronized with the video timeline and overlaid on the screen while the video is playing. The appeal of danmaku lies in its ability to create a “pseudo-synchronic” co-watching experience among viewers who are watching the same video (Johnson, 2013). This enhances both temporal and spatial contiguity for video viewers. While danmaku originally emerged in the context of animation videos, it has been introduced into the realm of MOOC learning and has seen a gradual rise in popularity. A study found that 76% of the surveyed students enjoyed using this innovative interaction tool in their learning experiences (Hu et al., 2017).

Many studies on danmaku focus on non-learning contexts such as communication perspectives (e.g., Liang, 2021), but there are also studies exploring danmaku in the context of online learning (e.g., Chen et al., 2019; Hu et al., 2017; Lee et al., 2015). Some studies highlight danmaku’s potential and its affordances in enhancing social interaction and interactivity among students in learning (e.g., Lee et al., 2015; Lin et al., 2018), while others delve into understanding the context and discourse of danmaku-based learning interactions (e.g., Li et al., 2022; Wu et al., 2018). Specifically, Wu et al. (2018) conducted a comprehensive comparison of user participation, language use, and knowledge sharing between danmaku and forums. They discovered that danmaku and forums facilitated knowledge sharing in complementary ways, with more explicit knowledge shared through danmaku and more tacit knowledge shared in forums. Although their findings provide a foundation for further research on knowledge sharing, their work primarily focused on the cognitive aspect of information exchange and did not delve into the socio-emotional aspects of danmaku and discussion-board interactions.

Methods

Historically, research in CSCL has focused on observing collaboration in a small-scale, controlled context (Wise & Schwarz, 2017). While this approach is effective for comparing interventions under controlled conditions and for examining various variables, it might overlook the study of collaborative learning in its most natural state. In the context of large-scale learning, it becomes valuable to also conduct descriptive analyses of informal environments in naturalistic settings. Such an approach expands our comprehension of collaboration within the context of utilizing tools primarily not designed for collaborative purposes.

Therefore, this analysis adopts a naturalistic mode of observation within an authentic MOOC lecture. The study applies a mixed-method approach in which quantitative content analysis as well as t-tests are used to showcase the overall pattern of the interaction, and in-depth qualitative analysis is used to closely examine the behavior of the interactions that take place. In this way, the process analysis changes from merely “coding and counting” categories to “exploring and understanding” how group interaction and meaning-making are achieved (Stahl et al., 2006).

Context and data

The analysis focuses on a video-based MOOC lecture about the quadratic function. This lecture was selected for analysis by first searching for all MOOCs in a danmaku portal in China, then filtering out those with too few danmaku messages or discussion-board posts (either unpopular or newly uploaded), and finally randomly choosing one from the remaining lectures. This is one lecture of an informal MOOC covering basic school math. It is an open course that anyone can take without paying a penny. However, since the instructor teaches the lecture based on content from a Grade 10 math textbook, it is assumed that the majority of the learners are Grade 10 students. However, there are learners that are not 10th graders. For instance, in the last example in Table 9, the learner was a Grade 12 student, and we identified another learner who claimed to be an adult.

The lecture is conducted in an asynchronous manner, and learners can watch the video and send danmaku or posts at any time. The length of the whole lecture video is around 24 min, and it consists of four parts (the definition, domain, and range of the quadratic function in addition to example exercises), each around 6 min. In the video, the instructor teaches in front of a whiteboard, as Fig. 1 shows. Both the lecture and the text generated by learners are in Chinese, but the examples and excerpts illustrated in the subsequent sections are translated into English for our readers’ understanding. Danmaku and the discussion board are displayed on one webpage of the MOOC portal, and learners usually need to scroll down the webpage to see the discussion board. The danmaku functionality can be switched on and off in the video player by the learners. Figure 2 indicates the layout of the three main zones on the MOOC’s interface.

Fig. 1
figure 1

Screenshot of danmaku in the video

Full size image
Fig. 2
figure 2

Layout of the MOOC’s interface

Full size image

In the period from October 2019 to September 2021, a total of 2435 danmaku messages involving 1371 learners and 739 discussion board posts sent by 269 learners were collected from this lecture. In addition, 32 posts sent by the instructor were identified on the discussion board, and there were no danmaku from the instructor. Since this study focuses on the interaction between learners, these 32 posts were excluded from the content analysis.

Analysis of mass participation

Regarding RQ1, we analyzed learners’ interaction between danmaku and the discussion board to explore how mass participation is accommodated accordingly. In this analysis, we reviewed all the danmaku messages and discussion board posts in a qualitative way to identify and compare the two tools’ respective patterns. This analysis focuses on interactional patterns in terms of the ways they are organized, interaction processes, and modes of learners’ participation in the discourse instead of the content of the learner-generated text. The content of the text will be analyzed in the subsequent sections using content analysis.

Quantitative content analysis

To answer RQ2, content analysis is harnessed in this study. Content analysis is frequently employed to examine transcripts of asynchronous computer-mediated discussions within CSCL environments (De Wever et al., 2006). The CoI model (Garrison et al., 2000) was used as the starting point for the quantitative content analysis. As argued by Stahl (2002), coding and counting pre-defined categories risk replacing the phenomenon of interest in the data. To mitigate this problem, a pilot analysis of 200 danmaku messages and discussion board posts randomly selected from the dataset was conducted before the coding scheme was made. The results showcased some socio-emotional interactions (related to disclosing feelings, giving compliments, greeting others, and finding a co-learning presence) and cognitive interactions (including asking or answering questions and sharing knowledge) among learners. Therefore, a coding scheme was developed on the basis of the template by Rourke et al. (1999) for accessing social presence in learning. Moreover, two cognitive categories, as identified in the pilot analysis, were also used. In addition, co-learning presence (Wei et al., 2012), which refers to the subjective experience of being together with others in a remote learning, was added to the scheme. In the subsequent process of coding all the messages and posts, it was found that some posts were left uncoded, since learners also tried to develop a deeper relationship beyond the course with peers; therefore, the scheme was updated by adding the category of “developing interpersonal relationships.” Table 1 presents the resultant coding scheme.

Table 1 Coding scheme for learners’ interactions using danmaku and the discussion board
Full size table

Since the text in danmaku and on the discussion board tends to be short, each complete danmaku or post serves as the unit of analysis. This also enhances the reliability of coding, since in the text data, danmaku and posts are naturally demarcated without the need for subjective segmentation by the coder.

Three persons (including the author and two assistants) familiar with danmaku participated in identifying and classifying the danmaku messages and discussion board posts. First, the author and one assistant worked independently as coders and coded all the data. Then, another assistant worked with the two coders to resolve any mismatches using the majority-rule approach. Between the two coders, a Cohen’s kappa coefficient of 0.84 and 0.88 was obtained for the classification of danmaku and posts, respectively. Both values showed substantial agreement (Viera & Garrett, 2005).

Based on the content analysis, the number of contributors (learners who sent at least one danmaku message or discussion board post) in each category of interaction in danmaku and on the discussion board was analyzed, and the mean (M) and standard deviation (SD) of the number of texts sent per contributor were calculated. This allowed a t-test to be conducted for the identified category to compare the means of the number of texts sent per contributor between danmaku and the discussion board. However, in each of these categories, only part of all contributors of danmaku and part of all contributors to the discussion board participated. Therefore, in the process of calculation, for those danmaku contributors and those discussion board contributors who did not participate in one category, that category was marked as “0” to indicate that they had not contributed any text related to it. Thus, through each category’s t-test result, we can compare how all contributors of danmaku and how all contributors to the discussion board participated in each category. This can extend the insights gained from the content analysis from the contributions level to the contributors level.

Qualitative content analysis

In the quantitative content analysis, the focus was on broader interactional categories rather than on individual interactional behaviors. Subsequently, a follow-up analysis guided by RQ3 was conducted with the aim of providing a more in-depth examination of the details within the interaction process. This analysis uses a qualitative method to highlight the outstanding interactional behavior across different content categories in both socio-emotional and cognitive interactions. Therefore, such a qualitative and descriptive analysis is implemented to complement the results of the quantitative content analysis to illustrate and explore the observed quantitative differences (Schwarz & Glassner, 2007).

Results

Analysis of mass participation

As illustrated in Table 2, danmaku and the discussion board have different patterns when it comes to the ways they are organized, interaction processes, and modes of learners’ participation in the discourse.

Table 2 Danmaku and the discussion board when it comes to the three aspects
Full size table

First, the two platforms exhibited different organizing methods. In Fig. 1, 20 danmaku messages can be identified in a single frame from the video. In the actual video, such messages are displayed in a continuous flow along with video playback. In contrast, the interaction on the discussion board is conducted through different threads (Fig. 3). There was a total of 98 threads on the discussion board, and on average, about 2.74 learners sent about 7.54 posts in each thread. As a result, these 269 learners’ participation was divided into small-group discussions accommodated in each thread on the discussion board.

Fig. 3
figure 3

An example of part of a thread on the discussion board

Full size image

Moreover, their interactions have quite differentiated processes. On the discussion board, thread-based discussions occur between individual participants, and it is very straightforward for interlocutors to respond directly to others by clicking the “reply to” button below each post or just sending a post directed at another user by adding “@user_id.” In danmaku, wherein there are no such “reply to” or “@user_id” functionalities, messages are usually sent to the crowd of mass participants instead of to a targeted addressee. However, a small number of direct replies to other danmaku were found. To this end, learners actively develop their own strategies, as Table 3 indicates.

Table 3 Replying strategies developed by learners in danmaku
Full size table

In terms of participation mode, learners’ engagement appears to be more scattered across different parts in the discourse in danmaku. For instance, one learner sent four danmaku messages, with one greeting the instructor at the start, one to find the co-presence of peers in the middle, and two thanking and praising the instructor at the end. On the contrary, on the discussion board, learners tended to participate by taking turns in one discussion. For example, one learner made three posts in a knowledge-sharing thread about the mapping relationship, first actively sharing his understanding, then providing further elaboration in response to another learner’s request, and finally commenting on one peer’s understanding. Notwithstanding such a general difference, how learners participate in danmaku versus on the discussion board may exhibit distinctive patterns in different categories of interaction, and this will be further analyzed in the following section about content analysis.

Quantitative content analysis

Table 4 presents the results of the coding, showing the categories identified, the number of their occurrences, and their main indicators. To showcase the distinction, the results for danmaku and the discussion board are given separately.

Table 4 Results of the content analysis of learners’ interaction
Full size table

As Table 4 indicates, learners’ interactions using danmaku and the discussion board share four out of all the categories. However, developing a relationship further is missing in danmaku, while cohesive responses and establishing the co-presence of peers were not identified among discussion board posts. In addition, more than 70% of the interaction in danmaku is in the socio-emotional dimension, while there is slightly more cognitive interaction than socio-emotional interaction on the discussion board.

In addition to their difference on the level of the content category of contributions, the discussion board and danmaku also differ on the level of contributors. Table 5. illustrates the result of the independent samples t-test for each category, and an alpha level of .05 is used. According to the t-test result, when it came to the mean number of text contributions sent per contributor in terms of knowledge sharing, the difference between all contributors using the discussion board (M = 0.859, SD = 0.633) and all contributors using danmaku (M = 0.129, SD = 0.393) was significant (t(1638) = −18.237, p < .001). Similarly, there were also significant differences regarding asking and answering questions, affective responses, and interactive responses between all contributors using the discussion board and all contributors using danmaku. Therefore, it was found that all contributors using the discussion board sent significantly more text per person than all contributors using danmaku in the four categories that both the discussion board and danmaku cover.

Table 5. Results of the independent samples t-tests in different categories to compare the participation of all contributors using the discussion board and all contributors using danmaku
Full size table

Qualitative content analysis

To illustrate the outstanding interactional behavior involved in the interaction process of the categories identified by the preceding analysis, excerpts of relevant messages and posts are used as examples in this section.

In danmaku, but not on the discussion board, the learners greeted the instructor at the beginning and end of the lecture. There were also greetings to other peers throughout the lecture (Table 6). In both danmaku and the discussion board, learners tended to send compliments to the instructor. On the discussion board, learners usually praised or thanked the instructor based on their overall learning experiences and gains after they finished the lecture. For example, one learner wrote, “I always got a score below 50 on my previous exams, but after learning the lecture here, I got 78 last week! Thx, my teacher!” In danmaku, compliments (Table 6) usually appeared (a) at the beginning of a lecture and displayed learners’ gratitude toward the instructor’s guidance in the previous lecture, (b) in the middle when the learner suddenly understood something they previously had a problem with or when the instructor provided a good solution, and (c) at the end of the lecture to thank the instructor for his good teaching in that lecture.

Table 6 Examples in danmaku of learners’ greetings and compliments toward the instructor 
Full size table

When the instructor was greeted and complimented through danmaku, these messages did not elicit any responses from the instructor (instructors’ messages are highlighted with special effects by the system). Instead, these messages were typically followed by similar messages produced by peers. Table 7 presents an excerpt of what could be characterized as a “co-monologue.”

Table 7 Example of a “co-monologue” message in danmaku
Full size table

On the discussion board, learners also sent compliments to the instructor, who responded to a small portion of such posts, but he respond much more often by liking such posts, as Fig. 4 indicates. The posts that were liked by the instructor are highlighted by the system with a special label, “the instructor liked this post.”

Fig. 4
figure 4

The instructor responding by liking the post

Full size image

The intention of developing deeper interpersonal relationships beyond the lecture with peers was seen only on the discussion board. Learners either shared their contact information or invited peers for subsequent learning outside the lecture. Table 8 presents an example.

Table 8 Posts on the discussion board to develop further relationships beyond the lecture
Full size table

In danmaku, learners also tried to establish the presence of other learners or indicate their own presence, whereas no such posts were found on the discussion board. Four ways to achieve this were identified in danmaku, and Table 9 presents examples.

Table 9 Different strategies for establishing and indicating co-learning presence in danmaku
Full size table

Moreover, these messages trying to find peer co-presence in danmaku could quickly receive tens of answers in a row, creating a “co-learning” presence. Table 10 presents an excerpt.

Table 10 Excerpt of answers demonstrating co-learning presence in danmaku
Full size table

Tacit knowledge accounted for the majority of knowledge sharing in danmaku in the form of self-disclosed experience and personal opinions about learning (e.g., “I always failed, because the key point here for me is not how to use the method of completing the square but when to use it.”). On the discussion board, however, explicit knowledge in the form of math knowledge (e.g., “The perfect square formula is: \({\left(a\pm b\right)}^{2}={a}^{2}\pm 2\text{ab}+{b}^{2}\)”) was mainly shared.

Danmaku formulated as questions regularly received a number of replies from peers, achieving a “co-answering” phenomenon, and Table 11 illustrates this in the excerpt. These answers tended to be rather brief, and they seemed more similar to quick tips, providing one facet of the complete information needed to answer the question. Taken together, however, the pieces might provide something that resembled a collective, full answer. In contrast to the discussion board, it was not convenient for questioners to ask for follow-up questions or demand further elaboration, as no “reply to” functionality was available. In the face of dozens of answers or two completely contradictory answers, it might therefore be challenging to choose the best or the right answer. In addition, not all requests could get the needed answer, and many obtained non-answer responses in the form of comments on the request itself or expressions repeating similar needs. Table 12 presents an example.

Table 11 Excerpts of the “co-answering” of a question in danmaku
Full size table
Table 12 Examples of non-answer responses to questions in danmaku
Full size table

When a question got a lot of such responses talking about similar need, as Table 13 shows, a “co-confusion” atmosphere could be constituted.

Table 13 Excerpt of “co-confusion” responses to a question in danmaku
Full size table

On the contrary, many questions received no or few responses on the discussion board. This means that not only were the learners’ needs for knowledge not satisfied but it could create a feeling of ‘lonely confusion’ that few peers were concerned with that issue. However, for some of the question-raising posts that did attract answers on the discussion board, several rounds of asking and answering, including follow-up questions and further elaboration on the initial answer, could be conducted. Table 14. presents the questions raised on the discussion board about an exercise given by the instructor (the same exercise as in Table 11). Furthermore, answers that were considered brilliant could be liked by others, and this might help learners choose the right or best answers if they felt unsure.

Table 14. Examples of rounds of asking and answering on the discussion board
Full size table

Sometimes, a synergy between danmaku and the discussion board could be observed. For example, two learners sent posts explaining the calculation procedure in detail on the discussion board as a response to questions raised in danmaku.

Discussion

In the subsequent parts of this section, we first discuss how mass participation was accommodated differently in danmaku and on the discussion board. Then, the findings in terms of socio-emotional interactions and then cognitive interactions are interpreted. Lastly, the different levels of joint interaction involved in the socio-emotional and cognitive interactions are discussed.

Mass participation versus small-group participation

The video timeline-synchronized feature in danmaku and the threaded design on the discussion board differentiate how they are organized. Furthermore, the “reply to” functionality plays a pivotal role in shaping the interaction process. In danmaku, the absence of this function helped to foster indirect crowd speaking, while on the discussion board, it encouraged direct communication between individuals within the thread. As a result, danmaku created a shared space for a large number of learners to interact indirectly, akin to stigmergic collaboration in large-scale contexts (Elliott, 2016). In contrast, the discussion board promoted small-scale person-to-person interactions consistent with many traditional CSCL practices. Given that learners can simultaneously benefit from both small-group and large-community environments (Eimler et al., 2016), the combination of danmaku and the discussion board in MOOCs has the potential to address diverse aspects of individual learners’ demands. In fact, the preceding analysis shows that, in this study, different aspects of learners’ needs (e.g., information retrieval and temporary while-learning connectedness) have been satisfied, at least to some extent.

The t-test for the category identified in the content analysis showcased that all contributors using the discussion board sent significantly more text per person than all contributors using danmaku. It can be noted that fewer learners participated in discussion board interaction and sent less text in total, but they generated significantly more text per person; while more learners participated in danmaku interaction and sent more text in total, they generated significantly less text per person. It shows that participation through danmaku was more scattered among a larger group of participants, whereas participation using the discussion board was more concentrated among fewer participants.

Socio-emotional interactions

The strategies employed in danmaku to reply to others reflected learners’ active efforts to engage in targeted interactions with their peers. These interactions were not one-sided but rather involved both the responder and the questioner, as demonstrated in the last example in Table 3. However, the danmaku’s design tends to be more crowd-oriented (Chen et al., 2015) and is therefore more suited for untargeted “idiolect” interactions within a crowd rather than for one-on-one dialogues on discussion boards. Compared with discussion board posts, which can contribute to interaction within smaller groups beyond the lecture, danmaku primarily serves as a means of temporary in-the-moment communication within a larger audience during the lecture. In line with this, attempts to foster personal relations between learners were observed on the discussion boards but not in danmaku. While danmaku messages usually do not target individual people, there was extensive use of greetings, phatics, and salutations. These elements served as indicators of cohesive responses, which are known to establish and maintain a sense of group commitment (Rourke et al., 1999).

In Chinese K–12 schools, it is customary to greet the teacher at the beginning and end of class, and learners may have continued this practice out of habit. In a pre-recorded MOOC, these greetings indicate learners’ efforts to simulate the classroom experience they are accustomed to, even though they understand that the instructor may not hear or respond to their greetings. In fact, learners engaged in parasocial interactions by offering such one-way greetings and compliments to the instructor. Parasocial interaction refers to an imagined social relationship, an illusion of friendship, and a simulated face-to-face interaction between the media user and the content producer (Jin & Park, 2009). Earlier research has identified parasocial interactions in the context of game video viewers who engage in imaginary conversations using danmaku (Wu et al., 2018). This suggests that learners may imagine and simulate their interactions with the instructor, much as in a traditional classroom environment. The “co-monologue” phenomenon identified in the analysis may have further encouraged parasocial interactions, as it created the illusion of shared experiences within the crowd rather than the isolated imagination of an individual learner. Learners using danmaku can perceive social connectedness with their peers and the instructor by employing greetings, phatics, and salutations.

The “co-learning” presence underscores the students’ desire to ensure that they are learning together with others in this MOOC. A sense of “co-learning” with other learners can be achieved by knowing that others are present and can be enhanced by receiving various responses from peers, as shown in Table 9. Consequently, learners’ feelings of isolation are likely to decrease when they sense the presence of others, making them more inclined to continue their learning, as previous studies have indicated that isolation is a major factor leading to attrition in online learning (Schaeffer & Konetes, 2010). The “co-learning” and “co-monologue” phenomena align with danmaku’s delivery of a “co-viewing” experience, a fundamental appeal identified in previous research on entertainment video consumption (Chen et al., 2015; Johnson, 2013). In contrast, no such “co-viewing” posts were found on the discussion board.

Collectively, all the behaviors described above represent lively socio-emotional interactions that contributed to the establishment of connectedness between peers, a sense of belonging to a “class” (community), and the development of friendly interpersonal connections. Given that learning in an online environment can be socially and emotionally demanding (Bakhtiar et al., 2018), fostering a favorable atmosphere through such interactions can enhance the overall learning experience for individual participants and facilitate group functioning.

Cognitive interactions

Messages in danmaku, which are more similar to spoken utterances (Liang, 2021), allowed learners to share tacit knowledge related to personal experiences and opinions in a casual chat-like environment. This finding contrasts with another study on knowledge sharing, which found that danmaku predominantly involved explicit knowledge sharing, while the discussion board exhibited more tacit knowledge sharing (Wu et al., 2018). The variation in findings may be attributed to the type of video content under analysis. In Wu et al.’s study, explicit knowledge sharing in danmaku was observed in a documentary video about feminism and the spherical radio telescope, subjects about which the average viewer had limited knowledge and experience, necessitating references to formal information. In contrast, in the MOOC on quadratic functions in this study, learners predominantly shared explicit knowledge on the discussion board, likely because posts which tend to be longer were better suited for sharing formal knowledge, including mathematical equations and symbols.

Within the collaborative environment of danmaku, a large number of participants contributed to a rapid accumulation of diverse responses to questions. These concise hints may not always have been the exact answers the questioner sought, but they could encourage learners to rethink their queries. Moreover, in addition to the informational aspect, many questions were framed as requests for help. When peers provided answers and “co-answering” was achieved, it carried socio-emotional significance, as this fosters a sense of connectedness and belonging to a considerate “learning community” for those seeking help. Such peer-help-seeking behavior holds substantial socio-emotional value in MOOC settings, where assistance from instructors is often limited (Van Der Zee et al., 2018), since seeking help socially from peers can reduce the dropout rate in MOOCs (Nelimarkka & Hellas, 2018). Even in cases where no answer is obtained, the atmosphere of “co-confusion” may alleviate the questioner’s anxiety and worries about difficulty, as it conveys the notion that they are not alone in their struggles, increasing the likelihood that they will continue their learning.

According to the cognitive presence of the CoI model, there are four phases: initiation, exploration, integration, and resolution. While the initiation (e.g., asking questions) and exploration (e.g., sharing personal narratives) phases were evident, the more advanced phases of integration and resolution were not identified in this study. This implies that learners primarily focused on information retrieval rather than engaging in higher-order cognitive activities, such as testing ideas or achieving resolution. Several factors may have contributed to this phenomenon. First, this was an introductory lecture, and advanced inquiry may not have been the learning objective for learners. Second, learners in this informal MOOC may have been motivated to acquire information primarily, as research using self-reported data has found that students typically use informal MOOCs to supplement their formal education with reliable sources of knowledge (Milligan & Littlejohn, 2017). Finally, the interactional processes in this context were self-directed by learners rather than moderated by an instructor or structured by collaborative scripts. In this case, the potential risk of overscripting (Dillenbourg, 2002) could be avoided since there was no coercion for the productivity of collaboration. Thus, activities could be conducted on the basis of learners’ real needs, and interactions could remain fun and rich, providing space for the lively socio-emotional communication found in the preceding section.

Different types of joint interaction

In the socio-emotional and cognitive interaction, various levels of joint interaction can be involved. Following the A3C framework (Jeong et al., 2017) and on the basis of the findings of this study, it can be argued that different types of interaction in terms of joint interaction level were conducted in danmaku and on the discussion board.

Those who passively observed danmaku and discussion board interactions, known as lurkers (learners who read others’ danmaku and posts without contributing their own), essentially engaged in “attendance.” They displayed minimal joint interaction by interacting with community- or group-created artifacts, such as danmaku and discussion threads while observing their peers’ interactions. These individuals can be considered part of the learning community, but they tended to take a more passive and marginal role. Learners who actively contributed by sending danmaku or discussion-board posts demonstrated “coordination.” This involved aligning their interactional behaviors with established norms, such as timing their danmaku contributions to match relevant learning content in the video or avoiding offensive content in their posts. Within this coordinated interaction, the learners transitioned from working in isolation to a more collective engagement, although they had not yet reached levels of cooperation or collaboration. Although learners did not explicitly divide tasks or determine specific contributions, they cooperated by providing various elements of the required contribution in a distributed manner. For example, as shown in Table 11, in the “co-answering” observed in danmaku, learners collectively worked toward a shared task of giving answers to a question raised by one of their peers. In the process of knowledge-based interaction through danmaku and the discussion board, collaboration emerged through knowledge co-construction, where learners built upon each other’s ideas, offering support or criticism. However, collaboration, necessitating extensive joint interaction in various forms to achieve shared understanding and collective responsibility, was observed only in a small portion of the knowledge-related discussions.

Engaging in asynchronous discussions is manageable in relatively small groups but becomes challenging in this MOOC interaction involving active contributions from a total of more than 1500 learners (Jeong et al., 2017). In this case, the learning community supported by danmaku displayed distinct patterns of interaction in an indirect or stigmergic manner (Elliott, 2016) compared with small groups but aligned more closely with large online knowledge communities such as Wikipedia (Jeong et al., 2017). While achieving joint activities as intimate as those in traditional small groups may be problematic or even infeasible in large-scale contexts, essential collaborative features, including shared goals, intentional coordination, and a certain level of group cognition beyond individual learning, could still be observed. Group work does not necessarily equal collaborative learning (Summers & Volet, 2010); however, as proposed by Cress et al. (2013), a process in a large-scale setting can be considered collaborative as long as it “fulfills the conditions that individuals act consciously following a common direction; that they take the perspective of the other participants into account; and that they contribute by building on the accomplishments of others” (p. 558). In this sense, part of the coordination and cooperation activities discussed earlier can be classified as collaborative processes.

Conclusions and implications

This study offers an analysis of interactions within danmaku and a discussion board in a genuine online learning setting. In CSCL, large-scale participation often involves a delicate balance between harnessing its benefits and addressing the challenges it brings. CSCL environments have historically emphasized the cognitive dimension of collaboration, sometimes overlooking the socio-emotional dimension. A scalable and sociable learning environment can be achieved when both the large and small scales of participation and both the two dimensions are taken into account and harmonized. In this regard, the roles of danmaku and the discussion board can be complementary. Danmaku is inclined toward socio-emotional aspects, characterized by its diversity, thanks to mass participation. In contrast, the discussion board is more focused on information retrieval and encourages intense interactions within small-thread groups. Danmaku primarily supports indirect interaction, while the discussion board serves as a platform for direct discussion. Danmaku fosters temporary connections among learners during the learning process, while the discussion board allows for the development of interpersonal relationships beyond the lecture. Furthermore, danmaku primarily facilitates the sharing of tacit knowledge, whereas the discussion board is more conducive to the sharing of explicit knowledge.

The combination of danmaku and the discussion board, as examined in this study, serves as a promising starting point, but it underscores the need for further well-designed socio-technical systems in this regard. Discussion boards, originally designed for small-group discussions, have demonstrated their continued utility in this evolving context, particularly when coupled with danmaku. This adaptability may apply to other tools originally created for traditional scenarios with fewer learners. By modifying and scaling them up to accommodate larger numbers of participants, valuable resources from established research in CSCL can be harnessed to reach a broader audience and have a more significant societal impact in light of emerging educational trends. The insights gained from this study offer practical guidance for educational design in the dynamic landscape of large-scale learning. Building on the analysis and discussions, the implications for educational design fall into two main categories: pedagogical design and environment and tool design.

First, in the pedagogical design of MOOCs, one aspect to consider is the instructor’s presence in learners’ interactions, as our analysis suggests learners’ genuine need for the instructor in MOOCs. Drawing from the concept of “teaching presence” (Anderson et al., 2001) within the CoI framework, in the lecture video, emphasis is placed on the instructor’s role as an instructional organizer, and the instructor also offered limited direct instruction by responding to learners’ questions on the discussion board. Therefore, an unfulfilled duty remains: the facilitation of discourse in learners’ joint interactions. However, achieving this with substantial increases in instructional staffing, as proposed in a recent study on danmaku (Li et al., 2022), may contradict MOOCs’ fundamental appeal of cost-effectiveness and, therefore, is not feasible. In fact, “teaching presence” is not equal to “teacher presence” and is the shared responsibility of the learning participants (Garrison, 2021). Learners have already taken part in “teaching presence” (e.g., the question-answering and knowledge-sharing interaction identified in the content analysis). As a result, learners themselves can also play the role of discourse facilitator if they are taught or prompted. The instructor can take this into consideration when designing the MOOC’s pedagogy to leverage the mass participation of learners to enhance the overall “teaching presence.”

Second, regarding environment and tool design, the current synergy between danmaku and the discussion board largely depends on the learners’ spontaneous usage. A more integrated design, combining danmaku and the discussion board is essential. To this end, deploying text-mining or natural language processing techniques in danmaku to automatically identify “hot spots” (e.g., the “co-confusion” identified in our analysis) and sending these as posts can seed the discussion board for more in-depth discussions (Shusterman et al., 2021). If applicable, the instructor can also use posts to elaborate on or comment on these identified “hot spots.” Conversely, natural language processing techniques can identify the knowledge-sharing posts on the discussion board that receive the most likes and display them around the relevant timeline in danmaku, providing learners with additional resources during the learning process.