Collaboration Technologies and Social Computing

This study proposes a method via which to support smooth rejoining in a face-to-face group chat for a participant (the leaver) after they have temporarily left, without interrupting the conversation. A previous study demonstrated that presenting a summary containing much of the recent conversational content to the leaver immediately before their return facilitates smoother rejoining for the leaver. However, because the conversation involves all participants, the issue of rejoining concerns not only the leaver but also those who continued the conversation (the conversers), who should be supported in continuing their conversation smoothly, without disruption. Therefore, in this study, we attempt to support smooth rejoining for both the leaver and the conversers. We propose a method of sharing common conventional information with both parties and discuss the effective types of information to share: a summary and keywords. An evaluation experiment suggested that sharing a summary enhances the leaver’s sense of being able to rejoin early and their ease of understanding the conversational content. Furthermore, sharing the summary reduced the conversers’ sense of the time required to explain the conversational content to the leaver. However, we could not confirm the effectiveness of sharing keywords in facilitating smooth rejoining. These findings suggest that when sharing common information before the leaver returns, it is important to share information that allows both parties to grasp the conversational content with the same level of detail and ease.

In educational settings, the trend of providing students with tablets for use during class is increasing. This shift is expected to encouraging encourage a transition toward active learning. However, fatigue hinders active learning. Interleaved learning has been proposed to reduce fatigue during learning sessions, but previous studies on interleaved learning did not control for the timing of task switching, possibly causing students to continue the same learning activity even when fatigued. In the current study, we hypothesized that switching tasks at the moment of fatigue can further help reduce fatigue. We developed a system that switches learning tasks based on fatigue levels (which are estimated by measuring leg movement) and examined changes in learner fatigue during its use. Evaluation experiment results indicated that the proposed method helped reduce fatigue, but the high frequency of task switching disrupted participant concentration immediately after each switch. In future work, we plan to investigate changes in learner fatigue when the system is used in group learning environments.

This study explored whether learners’ head movement and interhead distance could serve as behavioral cues related to learners’ internal states in small-group collaborative discussions: understanding, willingness to participate, and psychological safety. Eighteen science and engineering university students (in six groups of three) participated in discussions. A ceiling-mounted depth camera continuously measured their head movements during 20 min sessions. After their discussions, the participants watched the video recordings and rated their internal states per 1 min segment. Two groups were excluded due to head recognition errors caused by hand gestures, so the analysis focused on data from the four remaining groups. No consistent correlations were found across all learners or groups. However, some individuals and groups showed positive or negative correlations between internal states and behavioral indicators. Additionally, spikes in head movement often coincided with movements such as laughter, posture adjustments, and gesturing. These findings highlight the potential of bodily behavior as cues for understanding discussion dynamics. Future work will address measurement and environmental issues to elucidate these relationships better, promoting the reliable estimation of internal states from bodily behavior.

This study examines how two group formation methods—self-selected and random—affect students’ collaborative engagement, course learning outcomes (CLOs), and group formation preferences in a Software Engineering Professionalism course. A total of 247 students from a private university in Indonesia, across the Bandung and Purwokerto branch campuses, participated in the study. Data were gathered using the Collaborative Learning Engagement Scale (CLES) questionnaire, CLO scores based on project assessment rubrics, and preference questionnaires regarding group formation methods. The results indicated that the self-selected groups demonstrated significantly higher levels of behavioral and emotional engagement and performed better in achieving CLO 1 (professionalism). In contrast, the random groups had better scores on CLO 3 (professional communication) and maintained relatively stable cognitive engagement. Students showed a strong preference for the self-selected method, citing psychological comfort and work efficiency. However, the random grouping was beneficial in promoting adaptation and formal communication skills. These findings demonstrate that each method has distinct advantages: self-selected groups enhance comfort and collaboration, while random groups promote social resilience and professional skills. The study recommends adopting an adaptive or hybrid approach to group formation, aligned with specific learning objectives.

The widespread use of remote meetings enabled collaborative work without being co-located. However, remote meetings require participants to engage in multiple tasks simultaneously, such as browsing the document, texting in chat, and taking notes. These multitasking environments make it difficult to recognize who is about to speak or when to speak, which can lead to speech contention or awkward silence. To address this issue, we analyzed gaze behavior during three meeting conditions, in-person, audio, and video meetings, and two multitasking tasks, document browsing and chat replying, to find out the cues for smooth turn-taking. We analyzed the next speaker’s gaze direction just before taking a turn after speech contention or silence occurred. The result suggested that just before taking a turn after speech contention, the next speaker tends to gaze at the document or chat rather than at other participants, and after silence, the gaze direction is distributed. Based on these findings, we discuss the implications for facilitating smooth turn-taking in a multitasking videoconference.

This study presents a temporal analysis of user engagement, technology trends, and emotional dynamics on Stack Overflow across the pre-COVID, during-COVID, and post-COVID periods. Understanding these changes is crucial to identifying long-term shifts and enhancing digital engagement strategies in online developer communities. This is especially important during global disruptions like COVID-19, which reshape work patterns, collaboration, and community interactions. In this study, users were categorized by reputation and badges based on Stack Overflow’s reward system. For example, regular users were regarded as newer contributors, intermediate users are moderately active participants, and expert users are those who are highly trusted and recognized for their significant contributions and expertise. Our analysis revealed that engagement declined post-COVID, with regular users experiencing the steepest drop. Intermediate users showed signs of disengagement, strengthening the phenomenon of ‘leaky pipeline’, while expert users recovered the fastest in terms of engagement post-COVID. Technology trends shifted toward full-stack development and AI, and emotional analysis indicated high confusion and frustration during COVID, followed by increased admiration and approval after the pandemic, reflecting improved knowledge exchange. These findings provide actionable insights for fostering sustained participation and inclusivity in technical communities.

Stargazing often involves conversation about celestial objects, but perceptual differences such as visual acuity can cause misalignments in what participants see, making communication difficult. As a preliminary investigation, this study examined how visual acuity differences influence conversational behavior during collaborative stargazing. In a VR-based constellation search task, we compared pairs with matched and unmatched acuity. Although results were not statistically significant, consistent trends emerged: more clarification requests, higher question frequency, and longer response latency under acuity differences. These findings suggest that perceptual asymmetry may affect mutual understanding and point to the potential of conversation-based support systems.

Simulating collaborative learning is a critical yet challenging goal in educational technology. While recent Large Language Model (LLM) advancements show promise, existing approaches often rely on static error models and rigid dialogue control and are primarily designed as student-facing training tools. To address these limitations, we present an autonomous ‘zero-player’ multi-agent simulation platform, powered by GPT-4o, designed as a computational testbed for research. Our key contributions are a data-driven, probabilistic engine for modeling a realistic spectrum of student capabilities, and a fine-grained, consensus-driven dialogue protocol that fosters emergent, bottom-up collaboration. Qualitative evaluations demonstrate that our system generates sound, expert-aligned problem solutions and, critically, produces plausible collaborative dynamics, including peer-to-peer error identification and correction. Our work establishes a high-fidelity platform for studying the mechanisms of collaborative learning and lays the groundwork for future predictive tools to help educators optimize student grouping.

In distance project-based learning (PBL), reduced communication between teachers and learners makes it difficult to provide appropriate feedback, especially for inexperienced teaching assistants (coaches). This study proposes an LLM-based feedback generation method using learners’ activity reports and evaluates its effectiveness against coach-created feedback. An analysis of 216 informatics PBL activity reports showed that LLM-generated feedback outperformed coach-created feedback across feedback requirements, principles, and learning impact conditions. The findings suggest that LLM-based feedback generation can effectively support distance PBL.

Dialogue-style learning materials are superior to lecture-style learning materials in many aspects. However, creating dialogue-style learning materials places a significant burden on teachers. In this paper, we propose a generative AI system that applies Knowledge Graph-based Retrieval-Augmented Generation (KG-RAG) for creating dialogue-style learning materials. The initial evaluation suggested that the KG-RAG approach has the potential to generate consistent and educationally appropriate dialogues, even when high-quality existing teaching materials are not available. These findings indicate that KG-RAG may offer a promising direction for producing dialogue-style materials in resource-scarce contexts, potentially reducing instructor workload while maintaining pedagogical value.

Social capital (SC), defined as the network of relationships, trust, norms, and mutual support that facilitate cooperation within a community, plays a vital role in sustaining local economies, fostering community resilience, and preserving urban social cohesion. Despite its recognized importance, particularly within traditional commercial districts, quantitatively assessing SC remains a major challenge. This study introduces a dynamic analytical framework for evaluating SC in such districts by constructing interaction networks that capture relationships between individuals and retail establishments. A field experiment was conducted in a Tokyo shopping district by incorporating two interaction modalities: visitor-driven word-of-mouth (WOM) and store-driven communication. SC was measured using five questionnaire-based indicators encompassing necessity, attachment, trust, and social norms. Connection networks were constructed as weighted graphs based on interaction intensity. The results revealed that both WOM and store-driven communication significantly enhanced SC values, as confirmed through pre- and post-intervention statistical tests. A predictive model was developed to explore the SC and network structure relationships using eight network indicators. The model achieved reasonable accuracy and interpretability, with comparisons between complete and reduced feature sets providing insights into the trade-offs between generalization performance and explanatory power. This integrated approach of network analysis and machine learning offers a robust framework for understanding and quantifying SC dynamics. The proposed framework contributes methodologically and practically to the study of urban revitalization by capturing how specific interactions reshape network structures and community cohesion.

The aim of this study is to identify indicators of social exclusion in written texts expressing individuals’ impressions of content. Previous research has demonstrated that loneliness stemming from reduced interpersonal interaction negatively affects individuals’ health. Social exclusion is one of the reasons for causing loneliness. Nevertheless, many people find it difficult to openly disclose experiences of social exclusion. Accordingly, there is a need to develop methods that can find such exclusion without requiring explicit self-disclosure. In this study, we analyzed textual characteristics of reviews written by the experimental subjects. We divided the subjects into two groups by using cyber-ball task: a social excluded group and control group. Then, we asked the subjects in both group to read an assigned document and write a review of it. We analyzed these written review texts using natural language processing, such as sentiment and polarity analysis, to observe the emotional states of different experimental subjects. The findings indicate that texts produced by individuals experiencing social exclusion tend to exhibit heightened expressions of anger and a reduced use of neutral expressions.

In public spaces, communication among people with diverse values not only causes friction but also makes it hard to build relationships that bridge those differences. Simply expressing opinions is not enough to overcome divisions. To address this, we designed and implemented “Public Quest,” a communication game that frames a three-stage process within an RPG theme: (1) creative opinion expression through word games, (2) collaborative reconstruction of opinions via party formation, and (3) relationship building through face-to-face “Encounter” events. Through an exploration study, we observed that word games encouraged diverse expression. The “Name the Party” process led to the acceptance of a balanced range of opinions, while the “Encounter” event fosters positive dialogues and connections across differing views. This framing of communication as a shared quest supports not just creative expression, but also constructive engagement and relationship building in public spaces.

Traces left by people—such as footprints or signs of frequent passage—can serve as social cues that reveal the presence and activities of others in a space. This helps individuals decide where to look or what might be important, especially in unfamiliar environments or situations with many options. However, how such traces are interpreted and utilized during open-ended, unguided exploration remains poorly understood. Moreover, it is unclear how one’s own traces in such contexts influence cognition and exploratory behavior. In this study, we investigated in detail how traces left by others and one’s own past behavior are used during goal-free exploration and how they shape exploration patterns. Our results showed that traces left by others mainly functioned as cues for identifying exhibits with many or darker footprints as places worthy of attention, while the absence of traces sometimes stimulated curiosity. At the same time, however, they posed the risk of excessive expectations or overreliance. In contrast, seeing one’s own traces not only helped participants recall past behavior but also encouraged them to intentionally explore previously unvisited areas or try different routes.

People have discussions in idea generation. Discussion support systems are widely studied. For example, some systems visualize discussion contents by automatically generating transcriptions. Other systems provide participants with summarized discussion content. However, few studies focus on systems that extract discussion points that have not been sufficiently discussed. Identifying such insufficient discussion points allows participants to explore them more deeply. The discussion result will be sufficient, and make participants obtain well-considered ideas from the discussion results. This paper proposes a system that extracts topics that have not been sufficiently discussed, and provides them to participants for deeper exploration. The proposed system firstly clusters the utterance texts in a discussion. Then, the system extracts topics worth exploring deeply using cluster’s density. We hypothesize that one cluster expresses one discussion topic, and a low-density cluster indicates that the topic has not been discussed enough. The proposed system extracts low-density clusters as topic aspects worth deeper exploration and visualizes them using representative utterance texts. The discussion participants refer to these extracted topics for in-depth discussion, aiming to achieve more comprehensive results. We conducted evaluation experiments with participants. The experimental results showed that discussion results were characterized by high originality, validity, and efficiency when using the proposed system.

In this paper, we discuss the potential of using hackathons to promote critical discussions among participants from diverse backgrounds in the context of Artificial Intelligence. Our main objective for this work was to explore the viability of utilizing a one-day hackathon to bring together stakeholders, such as researchers, teachers, students and developers, to share their perspectives on AI literacy while collaboratively designing digital learning materials to promote AI literacy. In the current landscape, awareness and AI literacy have become essential. There is, however, no clear consensus on the definition of AI literacy, the needed target focus for teaching it, or how to design (digital) learning materials and curricula. Our findings show that the participants’ mixed backgrounds contributed to a meaningful discussion, and the hackathon outcomes were characterized as relevant and appropriate for teaching AI Literacy. We envision that this work contributes to the discussion about reaching a consensus on the definition of AI literacy, how to design (digital) learning materials and curricula, and who to include when creating these materials, by using a participatory design approach.

University student sports journalists often face time constraints during post-match interviews, as they are responsible for photography, live updates, and reporting with limited support. As a result, they have little time to take notes or prepare thoughtful questions, which often leads to vague or superficial interviews. Through interviews with student journalists and evaluations involving athletes, we found that questions focusing on specific plays and tactics were perceived more positively by both groups. To address this challenge, we developed a smartwatch-based system that allows journalists to flag important scenes during matches. These flags can later be reviewed using a prototype interface that displays both match videos and time-series information. We conducted two user studies to evaluate the system. The first study confirmed that journalists could successfully use the flagging function under real match conditions. The second showed that reviewing flagged scenes helped journalists formulate more play-specific and tactic-oriented questions. These findings suggest that the proposed system can support student sports journalists in improving the quality of post-match interviews, even under practical constraints.

This study introduces a structural framework for evaluating the quality of argumentative interaction in parliamentary debate. We proposed four hypotheses about rebuttal structures and defined corresponding features (Distance, Interval, Order, Rally). From a corpus of 20 English debate rounds with 1,573 ADUs and 679 rebuttal relations, we compared these features with human and LLM ratings. Regression analysis revealed a moderate correlation (r = 0.609), with Rally emerging as the most important predictor of interaction quality, followed by Distance and Interval, while Order showed limited explanatory power. To apply these insights in practice, we developed DebaTube, a visualization system that maps rebuttal structures to debate videos. A user study with experienced debaters confirmed that the system helps identify effective rebuttal patterns and improves exploration efficiency.