Collaboration Technologies and Social Computing

Inhaltsverzeichnis

1. Frontmatter

2. Full Papers

   1. Frontmatter

   2. HelaDepDet: A Novel Multi-class Classification Model for Detecting the Severity of Human Depression

      Y. H. P. P. Priyadarshana, Zilu Liang, Ian Piumarta

      Abstract

      Depression-driven suicide is a serious social problem. Early identification of depression is vital for the well-being of society. Clinical diagnosis of depression takes a significant amount of time and requires highly skilled medical staff, which greatly limits its accessibility. Social media analysis for depression detection is therefore a rapidly growing research area. However, most of the available methods can only detect the presence or absence of depression, not the severity of depression. On the other hand, a few recently developed models for depression severity detection have not been validated on large datasets due to fundamental issues such as data sparsity. In this study, we proposed a novel method based on confidence vectors for detecting the severity of depression. We evaluated our method using a large dataset consisting of more than 40,000 annotated statements extracted from multiple social network services. To our knowledge, this is the largest and most well-balanced dataset for depression severity classification to date. Preliminary results showed that our models outperformed the existing state-of-the-art models by 5%, achieving a micro-averaged F1 score of 66% for human depression severity detection.

   3. Analyzing Peer Influence in Ethical Judgment: Collaborative Ranking in a Case-Based Scenario

      Claudio Álvarez, Gustavo Zurita, Andrés Carvallo

      Abstract

      Peer influence is how an individual's beliefs, actions, and choices can be influenced by the opinions and behaviors of their peers. Peer influence can affect the moral behavior of individuals. In this study, we analyze peer influence in the context of case-based learning activity in ethics education. To conduct this type of activity, we introduce EthicRankings, a groupware environment that enables students to analyze an ethical case and reason about it by ranking the actors involved according to some ethical criteria. A study with a sample of 64 engineering students was conducted at a Latin American university to analyze peer influence from a dual standpoint in an activity comprising an individual response phase followed by a collaborative phase with anonymous chat interaction. Firstly, we determine how likely a student is to change their rankings in the collaborative phase when observing their peers’ rankings and interacting with them anonymously. Secondly, we compare positive, neutral, and negative sentiment variations in students’ written justifications for rankings before and after collaborating. Results show that students are highly likely to change their responses in the collaborative phase if their responses differ significantly from their peers’ in the individual phase. Also, sentiments in written ranking justifications vary in ways consistent with changes in ranking. The pedagogical implications of these findings are discussed.

   4. Exploring the Reciprocal Emotional Interaction Between Humans and Affective Technology Through Design Fictions and “Speculative Entanglement”

      Hong Yang, Ching-Yang Lin, Chung-Ching Huang, Ying-Yu Chen

      Abstract

      This paper explores how emotion recognition technology is perceived, understood, felt, and reimagined through a set of design fiction processes by making and re-making arrangements of complex relationships between technologies, practices, emotions, and everyday lives. Emotion plays a crucial part in how we interact with the world: it is ephemeral, contingent, and contextual. Technologies capturing human emotions trace how we feel from almost all perspectives of our lives. To start with our explorations, we first write design fiction and take them to workshops as a probe into complex relationships between emotion recognition and our everyday lives. Participants work in teams to create design fiction prototypes that expand and respond to the fictional worldbuilding. By analyzing how the entanglements of emotion recognition technology in human lives are carried through and transformed among multiple design fiction processes, we contribute to a set of design processes that uses design fiction as a probe for the next speculation, in this case, reconfigure emotional recognition among humans and technology in design fiction workshops.

   5. Citation Recommendation Chatbot for Professional Communities

      Alexander Tobias Neumann, Michal Slupczynski, Yue Yin, Chenyang Li, Stefan Decker

      Abstract

      In recent years, the proliferation of academic literature has made it increasingly challenging for researchers and professionals to discover relevant citations for their work. To address this issue, this paper presents CitBot, a novel Citation Recommendation Chatbot designed specifically for professional communities. We describe the design, development, and evaluation of CitBot focusing on its performance and usefulness. CitBot combines the citation context with document-level embeddings utilizing SPECTER to generate personalized citation recommendations based on the community’s research interests. The system is designed to seamlessly integrate with online professional platforms, providing users with citation suggestions in response to their queries. A user study was conducted to assess the chatbot’s performance, comparing it to other citation recommendation tools. The findings of the study, along with a discussion of CitBot’s benefits and limitations, are presented. By enhancing the citation discovery process, CitBot has the potential to improve the productivity of professional communities and transform the way researchers and practitioners access and engage with scientific knowledge.

   6. Differential Characteristics and Collaborative Interactions of Institutional and Personal Twitter Accounts in a Citizen Science Context

      Simon Krukowski, Fernando Martínez-Martínez, H. Ulrich Hoppe

      Abstract

      The analysis of Twitter data can help to better understand the interplay between scientific institutions, volunteers and other actors in a Citizen Science (CS) context. A first essential distinction has to be made between different user types such as organizations and individuals. We have applied and evaluated different machine learning approaches for this purpose. On this basis, we have analyzed networks based on different Twitter relations to characterize roles and interactions between different user types. Relations based on retweets, quotes, and replies capture the short term dynamics of on-going discussions. Our findings indicate that institutions are the main information sources, whereas personal users have an important role in active information spreading and dissemination through retweeting and quoting. Projecting the dynamic interactions onto a static network based on the follower relationship shows that pathways of dynamic information diffusion are largely determined by the static follower topology. These findings provide strategic information for managing CS-related discussions.

   7. Fairness in Socio-Technical Systems: A Case Study of Wikipedia

      Mir Saeed Damadi, Alan Davoust

      Abstract

      Wikipedia content is produced by a complex socio-technical systems (STS), and exhibits numerous biases, such as gender and cultural biases. We investigate how these biases relate to the concepts of algorithmic bias and fairness defined in the context of algorithmic systems. We systematically review 75 papers describing different types of bias in Wikipedia, which we classify and relate to established notions of harm and normative expectations of fairness as defined for machine learning-driven algorithmic systems. In addition, by analysing causal relationships between the observed phenomena, we demonstrate the complexity of the socio-technical processes causing harm.

   8. Students’ Generated Text Quality in a Narrative-Centered Learning Environment: Effects of Pre-Collaboration, Individual, and Chat-Interface Submissions

      Emily Theophilou, René Lobo-Quintero, Roberto Sánchez-Reina, Davinia Hernández-Leo

      Abstract

      Narrative-centered Learning Environments (NcLE) offer a powerful tool for enhancing students’ learning through interactive experiences. The integration of open-ended questions in NcLEs encourages students to express their thoughts, opinions, and ideas without any restrictions. While different submission formats are implemented in NcLE, little research has argued the importance of assessing their potential effects. This study, therefore, investigates how the text quality in writing varies based on different submission formats for open-ended submissions in a NcLE. We proceed to analyze three types of submission formats within a narrative scripts platform: a) formal individual, b) pre-collaboration, and c) a chatbot interface. For this study, data was collected in a randomized controlled trial that involved 311 secondary school students participating in a social media literacy workshop. The quality of the submissions was assessed using an automated text cohesion and coherence analysis tool (Coh-Metrix). Our results show that having formal submission pages for open-ended questions has an effect in students’ efforts to create submissions with higher writing quality, while submissions via chatbot interfaces tend to be brief and conversational in nature. The results presented in this study have practical implications for learning technology developers and educators, as they can guide decisions related to the creation, and integration of elements to promote writing quality in online learning environments.

   9. The Similarity of Virtual Meal of a Co-eating Agent Affects Human Participant

      Jui-Ying Wang, Tomoo Inoue

      Abstract

      In co-eating with real people, similar food consumption was found to benefit food intake and some subjective feelings. While co-eating agents have the potential to be caregivers or companions, the discussion of meal similarity between participants and agents is lacking. In this study, to achieve better social facilitation and the sense of eating together by a co-eating agent, we focused on the effects of meal similarity on eating amount and subjective feelings. We developed co-eating agents which can eat three types of food and conducted a laboratory-based artificial co-eating experiment. The results showed that participants perceived the meal similarity and the sense of eating together to be higher when the co-eating agent eats similar virtual food. In addition, a relationship was found between food tastiness and the difference of eating amount between the conditions. We propose that creating similar foods for co-eating agents can improve the feeling of togetherness in artificial co-eating and have the potential to facilitate eating when the food is preferred.

3. Work-in-Progress Papers

   1. Frontmatter

   2. Towards a Design Toolkit for Designing AR Interface with Head-Mounted Display for Close-Proximity Human-Robot Collaboration in Fabrication

      Yi Zhao, Lynn Masuda, Lian Loke, Dagmar Reinhardt

      Abstract

      Industry 5.0 puts forward clear requirements for improving the interactive experience of human-machine collaboration, but few design tools currently exist to assist designers to think through how to design human-robot interactions, with a focus on turn-taking, coordination and communication of the interactive intention flow. We present a design toolkit for designers to explore, specify, analyse and document the unfolding interaction between human(s) and robot(s) through a model of interaction where the roles of leader and follower shape the forms of collaborative behaviour. We illustrate our preliminary design toolkit in action and demonstrate how it can be used to aid in exploring new augmented reality applications in robotic fabrication.

   3. Competition or Cooperation: Classification in a VR Environment Based on Sensor Data

      Yoshiko Arima, Yuki Harada

      Abstract

      This study investigated activity during a joint Simon task in a VR environment to develop an analysis tool for collaboration teams using Metaverse software. In this pilot study, we distinguished between competition and cooperation during the task and explored the relationship between the classification model and performance on the joint Simon task. This study consisted of two phases: creation of the body movement classification model and adaptation of the model to the joint Simon task. In Phase 1, data from 6 participants (three pairs) were used to construct a machine-learning classification model. The other two participants provided test data for the model. Using random forest models, we classified two categories of pair movements: cooperation (synchronization) or competition. This model yielded an accuracy rate of 88.8% in classifying the test data. In Phase 2, as a case study, we applied this model to the joint Simon task. The results suggested that competition elicited better performance than cooperation. In conclusion, the classification model successfully distinguished subtle movements in the VR environment. This model could be used to analyze the state of pairs during collaborative tasks.

   4. Effect Analysis of Facemask on Emotions’ Communication Between Japanese People

      Yoko Nishihara, Azumi Inoue, Junjie Shan

      Abstract

      This paper reports the effect of a facemask on emotions’ communication between Japanese people through experiments with participants. In the experiments, we made participants in pairs and asked them to speak about their own episodes with emotion to another. The relationships of the pairs included “first-meeting” and “friends.” The participants picked out emotions on their own and spoke about their episodes with the emotions in Japanese. We separated the pairs of participants into two groups. Group A (18 pairs, 36 participants) was asked to communicate without a facemask, while Group B (18 pairs, 36 participants) was asked to converse with facemasks on. The results were compared between the two groups to reveal the effect of the facemask. From experimental results, we found that the effect of emotion conveyance would reduce for friend-pairs if they wore facemasks.

   5. Extracting User Daily Routine Activity Patterns from UWB Sensor Data in Indoor Environment

      Muhammed Rahim Fawad, Tessai Hayama

      Abstract

      In recent years, location-based technologies for ubiquitous environments have transformed individuals’ current location data into valuable assets. To establish advanced indoor location-based services, highly accurate positioning technology is required to precisely recognize and predict the movements of humans and objects. Hence, we proposed a method for extracting a user’s activity patterns from time-series ultra-wideband (UWB) tag data in indoor environments. The proposed method consists of three steps: 1) estimate the user’s stay regions from the user’s location history using UWB sensors attached to the user, 2) assign each stay region to significant indoor activities, and 3) mine the activity patterns and their characteristics of the user from the sequence of indoor activities. In our experiments, we confirmed that the proposed method performed better in recognizing activity regions indoors and that the activity patterns of each member in the laboratory were discovered in a practical environment using the proposed method.

   6. Semantic Network Analysis of a Learning Task Among Japanese Students of Psychology

      Vargas Meza Xanat, Shimojo Shigen, Yugo Hayashi

      Abstract

      The complexity of the learning process includes cognitive elements that are difficult to visualize in real time. Collaborative learning adds a social factor. In this study, we examined the case of Japanese university students in a psychology course who first worked individually and then in pairs to draw concept maps using a computer program. We focused on the Interactive Constructive Active/Passive (ICAP) framework of cognitive engagement through semantic and network analysis of the concept maps drawn by the students and their conversations. We drew network graphs to visualize the ICAP indicators across performance groups, uncovering that High Performers employed a wider diversity of nouns, keywords and connections related to the learning task than Low Performers. High Performers were more proactive and emotionally involved in the learning tasks. We confirmed that positive cognitive features are related to positive learning outcomes, providing recommendations for computer-supported collaborative learning (CSCL) systems according to students’ needs.

   7. Computational Analysis of the Belt and Road Initiative (BRI) Discourse on Indonesian Twitter

      Lotenna Nwana, Ugochukwu Onyepunuka, Mustafa Alassad, Nitin Agarwal

      Abstract

      The Belt and Road Initiative (BRI) is an ambitious development project to build road and sea infrastructure through parts of Asia, Europe, and Africa that will encourage international trade and development. However, since its launch, a major concern and central research theme has been the possibility of the initiative being a ‘debt trap’ and an opportunity for China to gain power over countries like Indonesia. Previous research adopting more qualitative approaches have identified negative reactions to the BRI in terms of China’s intentions. However, there is a gap in the extant research focusing on the systematic evaluation of the BRI discourse on social media leveraging computational methodologies, particularly content and network analysis. Understanding the structure of the BRI discourse network can reveal key information actors that are driving the propagation of information through the network. As such, we extracted 12,985 tweets from Twitter to understand the different topics being discussed about the BRI in Indonesia. Latent Dirichlet Allocation (LDA) topic model algorithm classified the tweets into topic groups and helped us understand each topic’s underlying theme. In addition, the user and user’s follower data was analyzed to understand the information flow network and identify the most influential users within the network. While some users speculated on China’s good intentions for Indonesia, others argued that Indonesia acts like foreign minions controlled by the Chinese government. Furthermore, we identified key information actors within the network that are important and well-positioned for the diffusion of information across the network.

   8. Conducting Morality and Emotion Analysis on Blog Discourse

      Stella Mbila-Uma, Ifeanyichukwu Umoga, Mustafa Alassad, Nitin Agarwal

      Abstract

      Social media has exploded in its usage, since the advancement of technology with individuals sharing their opinions and beliefs on its platform through text. This has led to the access of tons of text data that researchers can use to further understand human behaviour or user perspective. Emotion analysis is a common text analysis technique used to discover embedded emotion in content. For this research, we compare emotion analysis and morality assessment techniques in discovering embedded user opinions. Our results show that emotion analysis possesses limitations in its result. On the other hand, morality assessment provides a more comprehensive and accurate analysis of text data. For this research, we proposed that both methodologies are not mutually exclusive, and can complement each other to better understand the complexities of human communication and behavior.

   9. Gaze-Aware Social Interaction Techniques for Human-Robot Collaborative Shopping

      Masaya Iwasaki, Kosuke Ogawa, Tatsuyuki Kawamura, Hideyuki Nakanishi

      Abstract

      Robots that provide customer service in physical stores are being researched as a means of coexisting with people in everyday situations. One issue with such robots is that their suggestions are usually ignored and may not effectively promote purchasing behavior among customers. This paper aims to investigate whether customers are more likely to accept a robot’s suggestion by having the robot use personalized information that can enhance the shopping experience. To investigate the effectiveness of a robot’s suggestions, we conducted an experiment in a physical retail store in a real-world environment. The study aimed to encourage customers to pick up products by utilizing their posture information. As a result, the number of customers who picked up the product increased when the robot made suggestions when customers leaned forward to look at the product. This suggests that using a customer’s posture information to make suggestions can increase the likelihood of a customer accepting a robot’s proposal.

   10. Preliminary Study on Speaker Intimacy Focusing on Topic Continuity

       Takuto Miura, Hideaki Kanai

       Abstract

       In recent years, several studies have been conducted to estimate and quantify speaker intimacy based on information obtained from dialogue to improve the usability of dialogue systems. These studies have identified linguistic features (e.g., the presence or absence of honorifics) that contribute to estimating speaker intimacy. However, because these features can only superficially recognize intimacy, intimacy estimation based solely on these features may be less robust. Therefore, this study searched for features that could capture speaker intimacy without being limited by factors such as speaker attributes. In particular, we focused on topic continuity based on the similarity of topics between utterances rather than on-topic content. The results suggest that these features contribute to the estimation of speaker intimacy.

   11. Reshaping Group Life: A Transparent and Interpretable Reward Model to Enhance Fairness in Groups

       Jia-Wei Liang, Hao-Chuan Wang

       Abstract

       Groups can do better than individuals through two mechanisms: aggregation and synergy. Aggregation means bringing knowledge together, and synergy means increasing the effectiveness that comes about through joint action or cooperation. However, we usually measure a group’s effectiveness by productivity outcome but disregard the other critical aspects, specifically the experiences and sustainability of the team: does the group member feel fair? Without the sense of fairness, group members do not have a clear metric on how their contributions lead to rewards, and may gradually lose the motivation to engage and contribute. Groups can suffer both in terms of aggregation and synergy. Our goal in this work-in-progress paper is to formulate a user-interpretable and -transparent reward model to operationalize fairness in groups. We apply the model to design a workload tracking dashboard for group members to view and negotiate individual workloads transparently, and to improve fairness both in group procedure and outcome.

   12. Support How-To Instruction Following via Conversational Agent

       Qingxiaoyang Zhu, Yi-Chieh Lee, Hao-Chuan Wang

       Abstract

       People indispensably use instructions shared by one another to work on unfamiliar tasks in daily or professional life. However, personally shared tutorials are often based on personal experiences and represent a collective overview of past encounters, which can be misaligned with the specific work context during the re-enacting time. Drawing inspiration from the effective dynamics observed in conversational instruction-giving and -following between experts and novices, we propose a chatbot system that delivers archived how-to tutorials by providing necessary information in a just-in-time manner, tailored to the needs of the instruction-follower. Our aim is to transform unaided instruction reading activities into conversational instruction-following experiences. We implemented a chatbot system and evaluated it through a between subject study. The results demonstrate a promise of leveraging human-chatbot interaction to support actionable instruction-following.

   13. Social Pressure in Co-Manipulation: From Verification of Refrains in Communication During Fusion Avatar Manipulation

       Taichi Sono, Hirotaka Osawa

       Abstract

       In this study, we analyzed subjective comments on behavioral strategies that indicate negativity toward the partner during joint manipulation of the same avatar by participants who had never met each other before. We validated using the Tele-nininnbaori task: Two participants intervened with asymmetric information about the manipulation and the task for a single arm-shaped object and performed the task while communicating with each other through gestures. In previous studies, researchers have conducted validation between operators who have established some degree of relationship with each other. In this study, we designed a validation experiment in which two first-time participants were the experimenters. We obtained subjective comments on the behavioral strategies of the operators on the side who knew the task’s goal but had constraints on the operation in the form of post-experiment interviews, in which they expressed negativity toward the perceptions and actions of the other party. For the mentioned negation behavior strategies, two persons involved in the experiment made assignments to the classification items created by overlooking the strategies all participants took. As a result, we confirmed that the strategy of explicit denial by shaking the object sideways, which has been taken in many previous studies, was hardly used, confirming the occurrence of refrains due to social pressure of communication in multi-person avatar manipulation.

4. Correction to: Conducting Morality and Emotion Analysis on Blog Discourse

   Stella Mbila-Uma, Ifeanyichukwu Umoga, Mustafa Alassad, Nitin Agarwal

5. Backmatter