International Journal of Computer-Supported Collaborative Learning

Effective socially shared regulation of learning (SSRL) in computer-supported collaborative learning (CSCL) often fails to occur because students lack awareness of their peers’ and groups’ activities, thus leading to unsatisfactory learning outcomes. While both group awareness (GA) support and students’ self-regulated learning (SRL) levels are critical to CSCL, previous research has considered them separately, and investigation of their combined effects, especially on students’ SSRL, remains limited. Addressing these gaps, the present study conducted an 18-week experiment with a two-level factorial design to examine the main and interaction effects of GA support (present versus absent) and students’ SRL levels (high versus low) on perceived SSRL skills, observed SSRL behaviors, group task performance, and individual knowledge achievement. A total of 54 undergraduates enrolled in an Educational Research Methods course were randomly assigned to either a GA+ class (n = 28) or a GA− class (n = 26). The results revealed that: (1) GA support showed significant positive effects on all four measured variables, (2) SRL levels showed no significant main effects on these core measures, and (3) interaction analyses suggested that GA support substantially improved overall perceived SSRL skills among low-SRL students, whereas high-SRL students showed greater gains in the monitoring and adapting dimensions of observed SSRL behaviors. No interaction effect emerged for individual knowledge achievement. On the basis of these findings, several practical implications for facilitating successful collaborative learning are proposed.

Research on online problem-based learning—and computer-supported collaborative learning at large—has mostly focused on either the order of group members’ interactions (using time-oriented methods) or the co-occurrence of interactions (using network methods) within the same collaborative episode, while work on longitudinal dynamics has so far been lagging. In this study, we implement a novel method that combines the advantages of both approaches: the relational and temporal dimensions, which is temporal network analysis. Additionally, to capture changes at different temporal scales, we use sequence analysis and multilevel growth models to study how interactions and patterns unfold across time. Our results showed that students who used interactive socioemotional or regulated constructive patterns were more productive in terms of cognitive and knowledge productivity. Explicit group regulation was infrequent and emerged in response to challenges, questions, or disagreements, often with teacher support. Most groups settled into stable regulatory patterns early on, with limited change over time, and transitions—when they occurred—were usually between similar patterns. Our results also suggest that regulation does not naturally improve with time alone, underscoring the importance of early, targeted instructional support to foster more productive regulatory approaches.

Since the beginning of computer-supported collaborative learning (CSCL) research, collaborative writing has been playing a pivotal role as a tool for learning and knowledge construction. In the study presented here, we ask to what extent large language models may not only assist individuals in their writing processes but also serve as a collaboration partner. For this purpose, we analyzed the writing process of individuals supported by ChatGPT. We introduce the use of recurring n-grams as a means for textual uptake, that is, the extent and granularity with which human writers adopt and adapt artificial intelligence (AI)-generated text. On the basis of the overlaps between the ChatGPT output and participants’ final texts, we identified clusters of text reproducers, integrators, and reconstructors. Participants in these clusters differed not only in their subjective contributions and authorship but also in their prior use of ChatGPT and their affinity toward technology interaction. Referring to the conceptualization of interindividual interactions as uptake events, we suggest that n-grams are adequate means to analyze the uptake process in AI-supported human writing. Our findings show that AI-supported writing comprises distinct uptake patterns that differ systematically in the degree of textual reuse and perceived authorship, thereby revealing varying modes of engagement in human–AI co-writing, ranging from passive uptake of AI-generated text to more active and integrative forms of collaboration.

The formation of effective collaborative programming groups is vital for collaborative knowledge innovation. Previous research has predominantly examined the influence of group composition approaches from a computational perspective, yet there remains a limited resolution of their real-world educational impacts. This study offers empirical insights into the effects of homogeneous versus heterogeneous groups on student performance within collaborative programming contexts. The group composition system was established using a genetic algorithm, with the inclusion of socio-emotional competence, learning styles, and academic achievement. A total of N = 478 students aged between 13 and 15-years-old voluntarily participated in the study and were divided into 42 heterogeneous groups (n = 166), 40 homogeneous groups (n = 163), and 36 random groups (n = 149) with a group size of four. All participants were subjected to identical pedagogical conditions under a double-blinded study design. Collaborative programming performance was assessed both summatively and formatively, incorporating multi-source evidence from teacher observations, student self-reports, and peer evaluation scores. The results indicate that heterogeneous groups notably outperform homogeneous groups and random groups across most measurements. Implications for implementing collaborative programming in real-world classroom settings are provided.

As smart technologies become part of daily life, families face new opportunities and challenges in learning together. This paper introduces FamiData Hub, a speculative computer-supported collaborative learning (CSCL) prototype that supports families in building critical data literacy within smart homes. Through workshops with 17 families, the study explores how collaborative learning emerges through interaction, storytelling, and shared problem-solving, with family roles shifting dynamically. The findings challenge traditional adult-to-child teaching models, proposing instead a multidirectional learning space where anyone—including children and digital tools—can be the “more knowledgeable other.” The study highlights the value of family centered, socially embedded approaches to critical data literacy and offers insights for designing intergenerational CSCL systems to foster critical data literacy.

Mutual engagement, the dynamic process through which collaborators reciprocally take up and sustain one another’s ideas and actions, is crucial to collaborative problem solving (CPS). However, existing research has yet to fully specify concepts or methodologies needed to capture these dynamic characteristics. This gap highlights the need to examine how these patterns evolve across different CPS phases to inform more sophisticated instructional strategies that enhance collaborative learning. This exploratory study integrates multimodal and content analyses to examine phase-sensitive patterns of mutual engagement in small teams. In total, 28 college students participated in video-recorded CPS activities across four distinct phases. The findings revealed that high-performing teams displayed structurally complete elaborative sequences, in which invitations to contribute were taken up, elaborated, and reciprocated. Low-performing teams, by contrast, exhibited fragmented sequences that failed to return to elaboration. These interactional differences co-occurred with distinct multimodal signatures. High-performing teams exhibited greater interest, less frequent neutral emotions, and early posture synchrony, patterns that were especially pronounced during the ideation phase. Conversely, low-performing teams showed lower interest, persistent neutral emotions, and late, reactive posture synchrony. The findings elucidate the socio-cognitive characteristics of mutual engagement and demonstrate the potential for integrating emotional and behavioral indicators for a richer understanding. These insights can inform the design of instructional scaffolding and phase-sensitive support systems to enhance successful collaborative learning.

Productive collaborative discourse requires students to continuously advance ideas, often through the creation, modification, and integration of digital artifacts in a communal space. Without these processes, ideas remain isolated, fragmented, and unable to advance shared understanding. To support such discourse processes, this study proposes a knowledge synthesis (KS) intervention to facilitate a process of creating knowledge syntheses from ideas represented in digital artifacts and then leveraging these knowledge syntheses, represented in new digital artifacts, to deepen student collaboration. To examine the enactment of this intervention in a graduate-level course, we asked: What were the key characteristics of students’ knowledge synthesis artifacts? How did student groups use the synthesis artifacts during their discourse? To what extent did the synthesis artifacts facilitate collaborative discourse? We analyzed multiple data sources—including student-created synthesis artifacts, perception data, classroom video recordings, and co-constructed group artifacts—using a combination of descriptive, content, and interaction analyses. Findings revealed diverse approaches to knowledge synthesis and showed that synthesis artifacts facilitated discourse progression, fostered a range of knowledge practices, and supported the evolution of group artifacts. By promoting knowledge synthesis and examining its role in collaborative discourse, this study contributes to computer-supported collaborative learning (CSCL) by advancing the theoretical understanding of knowledge synthesis and offering pedagogical strategies for supporting this practice in classrooms.

In collaborative learning game environments where competition and collaboration coexist, conflicts among students are not uncommon. While conflicts of ideas and opinions are prevalent during collaborative learning, they are often perceived as elements to be avoided. One of the main concerns about conflict is its ability to trigger negative emotions, such as anger, which can compromise effective peer interaction, collaborative learning, and, in turn, diminish the quality of group discussions. However, this raises the question of whether anger always negatively affects collaborative learning. Most studies on negative emotions are related to test anxiety or boredom, while the impacts of emotions such as anger on learning are less explored. Especially within computer-supported collaborative learning (CSCL), there is limited research on how anger impacts students’ collaborative activities and learning. To address these issues, this paper aims to explore the potential relationship between anger and its impact on students’ collaborative discourse in a hybrid game-based simulation. Our findings suggest that anger has the potential to facilitate diverse and productive collaborative discussions. Students, driven by their anger, delved deeper into game mechanics, linked concepts to real-life situations, and employed various forms of logical reasoning to substantiate their opinions. However, the moment a student exhibited “tilting“ behavior, the quality of their collaborative discussions plummeted. Our findings provide important preliminary insights into the concept of “tilting” within immersive collaborative learning games and how it may manifest; they also offer guidance on the timing for educators’ intervention in collaborative discourse when anger arises among students.

Background: Online hate speech on social networks and the Internet is an increasingly pervasive phenomenon to which both children and adolescents are exposed. Objective: Our study’s main objective was to ascertain whether collective intelligence can improve their handling of hate speech. Methods: We conducted the study on the Collective Learning platform, comparing results between three groups of Spanish adolescents aged 15–16 years. The groups were of different sizes: one large group (G1, n = 123) and two smaller groups (G2, n = 18; G3, n = 23). Results: The experiment showed that the conditions for the emergence of collective intelligence were met within the large group (G1) but not in the two small groups (G2 and G3). The large group, as a collective, acquired capacities to deal with hate speech; however, this did not occur in the two smaller groups. Conclusions: Our study explains how the emergence of collective intelligence in online environments helps group members acquire a series of competencies. In particular, collective intelligence can help adolescents learn to deal with hate speech.

The increasing availability of multimodal sensing technologies has opened new avenues for studying human interactions. However, there remains a lack of systematic synthesis regarding which multimodal metrics are most predictive of productive collaborations. This study addresses this gap by conducting a systematic literature review of 163 studies published since 2000. Grounded in the theoretical framework of multimodal collaboration analytics (MMCA; Schneider et al., 2022), we examine how different data modalities—verbal, gaze, body, head, log, and physiological—are used to assess collaboration. Using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) framework (Liberati et al., 2009), we categorize studies on the basis of the types of collaborative indicators, the metrics extracted from multimodal data, and the methods used to establish relationships between them. We find several gaps, including an over-representation of lab-based studies with small sample sizes, reliance on simplistic individual or group synchrony metrics, and a lack of standard indicators for collaboration. We discuss related Grand Challenges for MMCA, including scaling up research through field-based studies, developing interpretable models that contribute to theory, computing sophisticated sensor-based metrics that better capture the temporal dynamics of interaction, and designing interventions that support collaboration using fine-grained, high frequency sensor data.

Collaborative learning deepens understanding by elaborating knowledge and facilitating memory-related information processing through interactions with others. In computer-supported collaborative learning (CSCL), mechanisms identified in collaborative learning are scaffolded through tools such as group awareness and scripted collaboration. While collaborative learning is considered effective, it remains unclear how older adults learn in collaborative environments using concept maps, and how cognitive decline may hinder their learning. Therefore, this study investigates differences between younger and older adults in collaborative learning with concept maps, focusing on learning performance, concept map performance, and the collaborative learning process. Learning performance was assessed using test scores, concept map performance through concept map evaluations as a tool for externalizing knowledge, and the collaborative process using the Interactive-Constructive-Active-Passive (ICAP) framework, which captures cognitive engagement. Results showed that younger adults had higher learning performance than older adults, while older adults showed no significant improvement, indicating a lack of learning gain. Similarly, younger adults outperformed older adults in concept map performance, and no improvement was observed in older adults for concept map scores. This suggests that older adults found it more difficult to elaborate knowledge, such as integrating new information. Regarding the collaborative learning process, younger adults were more likely to engage at the active, while older adults showed higher engagement at the constructive and interactive levels. Epistemic network analysis (ENA) revealed stronger connections between constructive and interactive behaviors in younger adults, and between active and interactive behaviors in older adults. These findings suggest that while younger adults progressively deepen their engagement during collaborative learning, older adults may require the reactivation of memory to engage in elaboration. These results offer insights into designing effective CSCL environments tailored to the learning needs of older adults.

Dialogic education is largely advocated as a means to promote dialogue and reduce polarization. Chatbots based on large language models (LLMs) carry the potential to scale dialogic education by serving as conversation partners and sustaining a dialogic space on various topics. They combine human-like conversational abilities with machine patience. To explore this potential, we fine-tuned an LLM-based chatbot, LlamaLo, using a corpus of productive discussions. We analyzed ten discussions with LlamaLo on contentious topics, such as liberalism and cultural appropriation. Our findings show that LlamaLo effectively opens dialogic spaces by questioning interlocutors’ assumptions, presenting alternative perspectives, and providing relevant knowledge. However, challenges, such as negative tone and bias, could undermine the dialogic space and should be addressed computationally and pedagogically. We conclude that dedicated LLM-based chatbots have the potential for enhancing dialogic education and enabling seamless scripting responsive to real-time needs.