Autonomous vehicles and robotic systems are increasingly used to perform operations in environments that bear potential risks to humans (e.g. areas affected by natural disasters, warfare, or planetary exploration). One source of danger is the contamination with hazardous substances. In order to improve situational awareness and planning, such substances must be detected using the sensors of the autonomous system. However, training a supervised machine learning model to detect different substances requires a labelled dataset with all potential substances to be known in advance, which is often impracticable. A possible solution for this is to pose an anomaly detection problem where an unsupervised algorithm detects suspicious substances that differ from the normal operation environment. In this paper we propose SpectrAE, a convolutional autoencoder-based system that processes multispectral imaging data (covering visible to near-infrared ranges) to identify surface anomalies on roads. Unlike traditional detection methods such as gas chromatography and physical sampling that risk contamination and cause operational delays, or laser-based remote sensing techniques that require pre-localisation of potential hot spots, our approach offers near real-time detection capabilities without prior knowledge of specific hazardous substances. The system is trained exclusively on normal road conditions and identifies potential hazards through localised reconstruction loss patterns, generating Areas of Interest for further investigation. Our contributions include a robust end-to-end detection pipeline, comprehensive evaluation of system performance, and a roadmap for future development in this emerging intersection of autonomous systems and crisis response technologies.

Graph databases are used across several domains due to the intuitive structure of graphs. They are well-suited for storing document collections together with their interlinkages through metadata and annotations. Yet, querying such graphs requires database experts' involvement for query formulation, reducing accessibility to nonexperts. To address this issue, we present Graph Detective, a web interface that provides an intuitive entry point for graph data exploration, where users can create queries visually with little effort, eliminating the need for expertise in query writing. After processing, the resulting query output (a graph) is then rendered in an interactive 3D visualization. This visualization allows the analysis of structural traits of the resulting graph data, exploiting the documents and metadata interlinkage. Our user evaluation revealed that even individuals inexperienced with graph databases or graph data, in general, could satisfactorily access the graph data through our interface. Furthermore, experienced participants commented that our interface was more efficient than writing explicit queries in graph database query language. Interested users can find the code openly accessible on GitHub1.

Abstract Web search is a crucial technology for the digital economy. Dominated by a few gatekeepers focused on commercial success, however, web publishers have to optimize their content for these gatekeepers, resulting in a closed ecosystem of search engines as well as the risk of publishers sacrificing quality. To encourage an open search ecosystem and offer users genuine choice among alternative search engines, we propose the development of an Open Web Index (OWI). We outline six core principles for developing and maintaining an open index, based on open data principles, legal compliance, and collaborative technology development. The combination of an open index with what we call declarative search engines will facilitate the development of vertical search engines and innovative web data products (including, e.g., large language models), enabling a fair and open information space. This framework underpins the EU‐funded project OpenWebSearch.EU, marking the first step towards realizing an Open Web Index.

Identifying vital nodes in complex networks is pivotal across various research domains such as social network analysis, epidemiology, and physics, with centrality measures being commonly employed. Despite growing privacy concerns, its impact on vital node identification, especially in networks with sensitive data like Bluetooth-based contact networks, remains underexplored. Our study assesses centrality measures' efficacy under constrained privacy settings, where only limited neighbor information is accessible. Through simulations, we pinpoint algorithms optimal for privacy-sensitive node vitality estimation, emphasizing the influence of network characteristics and algorithmic traits like multi-aggregation. This work enhances the understanding of privacy-centric methods in complex network analysis.

Software version control platforms, such as GitHub, host millions of open-source software projects.Due to their diversity, these projects are an appealing realm for discovering software trends.In our work, we seek to quantify synergy between software projects by connecting them via their similar as well as different software features.Our approach is based on the Literature-Based-Discovery (LBD), originally developed to uncover implicit knowledge in scientific literature databases by linking them through transitive connections.We tested our approach by conducting experiments on 13,264 GitHub (opensource) Python projects.Evaluation, based on human ratings of a subset of 90 project pairs, shows that our developed models are capable of identifying potential synergy between software projects by solely relying on their short descriptions (i.e.readme files).

Establishing small learning groups in online courses is a possible way to foster collaborative knowledge building in an engaging and effective learning community. To enable group activities it is not enough to design collaborative tasks and to provide collaboration tools for online scenarios. Collaboration in such learning groups is prone to fail or even not to be initiated without explicit guidance. In the target situations, interventions and guiding mechanisms have to scale with a growing number of course participants. To achieve this under privacy constraints, we aim at identifying target indicators for well-functioning group work that do not rely on any kind of information about individual learners.

In this integrated study of dynamics in MOOCs discussion forums, we analyze the interplay of temporal patterns, discussion content, and the social structure emerging from the communication using mixed methods. A special focus is on the yet under-explored aspect of time dynamics and influence of the course structure on forum participation. Our analyses show dependencies between the course structure (video opening time and assignment deadlines) and the over-all forum activity whereas such a clear link could only be partially observed considering the discussion content. For analyzing the social dimension we apply role modeling techniques from social network analysis. While the types of user roles based on connection patterns are relatively stable over time, the high fluctuation of active contributors lead to frequent changes from active to passive roles during the course. However, while most users do not create many social connections they can play an important role in the content dimension triggering discussions on the course subject. Finally, we show that forum activity level can be predicted one week in advance based on the course structure, forum activity history and attributes of the communication network which enables identification of periods when increased tutor supports in the forum is necessary.

Organizations conduct series of face-to-face meetings aiming to improve work practices. In these meetings, participants from different backgrounds collaboratively design artifacts, such as knowledge or process maps. Such meetings are orchestrated and carried out by facilitators and the success of the meetings almost solely depends on the experience of the facilitators. Previous research has mainly focused on approaches that support facilitators and participants in the upfront planning of such events. There is however, little guidance for facilitators and participants once a meeting has started. One critical aspect -- among others -- is that during a meeting, the facilitator and participants need to decide for how long the iterative process of discussion and design should continue. We argue that we can provide support for such decisions based on the evolution of artifacts collaboratively created during such meetings. This paper presents a multi-level, multi-method analysis of artifacts based on experts' observations in combination with network analytics. We study the use of automated analytics to assess the evolution of collaboratively created artifacts and to indicate maturity and established consensus of the collaborative practice. We propose a computational approach to support facilitators and participants in deciding when to stop face-to-face meetings.

This paper describes the analysis of the social and semantic structure of discussion forums in massive open online courses (MOOCs) in terms of information exchange and user roles. To that end, we analyse a network of forum users based on information-giving relations extracted from the forum data. Connection patterns that appear in the information exchange network of forum users are used to define specific user roles in a social context. Semantic roles are derived by identifying thematic areas in which an actor seeks for information (problem areas) and the areas of interest in which an actor provides information to others (expertise). The interplay of social and semantic roles is analysed using a socio-semantic blockmodelling approach. The results show that social and semantic roles are not strongly interdependent. This indicates that communication patterns and interests of users develop simultaneously only to a moderate extent. In addition to the case study, the methodological contribution is in combining traditional blockmodelling with semantic information to characterise participant roles.

There is an underexploited potential in enhancing massive online learning courses through small-group learning activities. Size and diversity allow for optimizing group composition in small-group tasks. The purpose of this paper was to investigate how groups formed based on learner behavior affect productivity of students in a small-group task. Students classified as high, average and low were randomly assigned to homogeneous or heterogeneous groups. Results indicate that overall, heterogeneous groups were either similarly or a bit more productive than homogeneous groups. Yet, we found that homogeneous groups classified as high-level were as or more than heterogeneous groups. However, heterogeneous groups were still more productive than homogeneous-average and homogeneous-low groups suggesting heterogeneous groups are the best choice for the entire community. Students classified as low-level were more productive in homogeneous groups, suggesting that grouping less active students together, makes social loafing more difficult and students participate more.

In this demo paper we present a new visualization technique for dynamic networks. It displays the time slices of the dynamic network using two dimensional graph layouting algorithms and stacks these in the third dimension to show the development over time. The visualization ensures that the same node always has the same position in each time slice so that it is easy to follow its development. It also allows filtering data and influencing node appearance based on properties. Additionally we offer a two dimensional comparison view for two time slices which highlights changes in graph structure and (if available) in measures of nodes. The presented visualization technique is implemented using web technology and is available in a web-based analytics workbench. We demonstrate the benefits of these techniques by an analysis of a data set from a learning community.

This work explores methods to investigate the structure of knowledge exchange in discussion forums in massive open online courses (MOOCs) explicitly taking into account changing patterns over time. Various aspects of forum analysis combining different approaches are exemplified with a case of forum discussions from a Coursera MOOC.

Technology supported learning is nowadays often based on heterogeneous environments that encompass not only one application and also possibly involve different devices. This creates specific challenges for data analysis and thus for learning analytics. In this paper, we propose a framework to create reusable learning analytics components that are portable to different target platforms. In this approach, the logic of each analysis component is specified in a separate web-based visual environment (or "workbench") from where it is later exported to the target environments form of a gadget-based dashboard. We demonstrate this mechanism in the context of the Go-Lab portal for accessing remote laboratories in STEM learning scenarios. An example shows how such analytics gadgets can be used to support collaboration inside the classroom.

This paper presents an analysis of resource access patterns in a recently conducted master level university course. The specialty of the course was that it followed a new teaching approach by providing additional learning resources such as wikis, self-tests and videos. To gain deeper insights into the usage of the provided learning material we have built dynamic bipartite student -- resource networks based on event logs of resource access. These networks are analysed using methods adapted from social network analysis. In particular we uncover bipartite clusters of students and resources in those networks and propose a method to identify patterns and traces of their evolution over time.

In this paper we introduce the concept of a web-based analytics workbench to support researchers of social networks in their analytic processes. Making explicit these processes allows for sound design, re-use, and automated execution using an authoring system for visual representations of these analytic workflows. The workbench is implemented according to a flexible technical framework in which external and newly-defined analytic components can be integrated and used in conjunction with other analytic components. As a showcase we discuss a complex analytic process.

This paper provides an introduction to the scientometric method of main path analysis and its application to detecting idea flows in an online learning community using data from Wikiversity. We see this as a step forward in adapting and adopting network analysis techniques for analyzing the evolution of artifacts in knowledge building communities. The analysis steps are presented in detail including the description of a tool environment ("workbench") designed for flexible use by non-computer experts. Through the definition of directed acyclic graphs the meaningful interconnectedness of learning resources is made accessible to analysis in consideration of the temporal sequence of their creation during a collaborative process. The potential of the method is elaborated for analyzing the overall learning process of a community as well as the individual contributions of the participants.

This paper introduces an agglomerative method for detecting cohesive subgroups in networks based on geodesic distance. The algorithm starts with a set of nodes as "seed". Beginning with the seed nodes as initial clusters, the clusters grow by incorporating more nodes successively based on minimal average distance to the current members of the cluster as a criterion for cluster extension. This approach is combined with an optimization step to achieve high quality performance on subgroup detection. The resulting method for detecting discrepant cohesive subgroups has been tested on artificial benchmark graphs as well as real-world networks.