The classification of short texts is a common subtask in Information Retrieval (IR). Recent advances in graph machine learning have attracted interest in graph-based approaches for low resource scenarios, showing promise in such settings. However, existing methods face limitations such as not accounting for different meanings of the same words or constraints from transductive approaches. We propose an approach which constructs text graphs entirely based on tokens obtained through pre-trained language models (PLMs). By applying a PLM to tokenize and embed the texts when creating the graph(-nodes), our method captures contextual and semantic information, overcomes vocabulary constraints, and allows for context-dependent word meanings. Our approach also makes classification more efficient with reduced parameters compared to classical PLM fine-tuning, resulting in more robust training with few samples. Experimental results demonstrate how our method consistently achieves higher scores or on-par performance compared to existing methods, presenting an advancement in graph-based text classification techniques. We also find that our results are on-par with prompting an LLM but with lower computational costs. To support reproducibility of our work we make all implementations publicly available to the community.

The widespread use of social media has highlighted potential negative impacts on society and individuals, largely driven by recommendation algorithms that shape user behavior and social dynamics. Understanding these algorithms’ impact is essential but challenging due to the complex, distributed nature of social media networks as well as limited access to real-world data and in particular recommendations, usually not reported. This study proposes to use academic social networks as a proxy for investigating recommendation systems in social media. By employing Graph Neural Networks (GNNs), we develop a model that separates the prediction of academic infosphere (in which a recommender can play a main role) from user behavior prediction, allowing us to simulate recommender-generated infospheres and assess different recommenders’ impact on the model’s performance in predicting future co-authorships. Our approach aims to improve our understanding of recommendation systems’ roles and social networks modeling. To support the reproducibility of our work we publicly make available our implementations: https://github.com/DimNeuroLab/academic_network_project

Pre-training of neural networks has recently revolutionized the field of Natural Language Processing (NLP) and has before demonstrated its effectiveness in computer vision. At the same time, advances around the detection of fake news were mainly driven by the context-based paradigm, where different types of signals (e.g. from social media) form graph-like structures that hold contextual information apart from the news article to classify. We propose to merge these two developments by applying pre-training of Graph Neural Networks (GNNs) in the domain of context-based fake news detection. Our experiments provide an evaluation of different pre-training strategies for graph-based misinformation detection and demonstrate that transfer learning does currently not lead to significant improvements over training a model from scratch in the domain. We argue that a major current issue is the lack of suitable large-scale resources that can be used for pre-training.

Social Media Artificial Intelligence algorithms provide users with engaging and personalized content. Yet, the personalization of algorithms may have a negative impact on users who lack AI literacy. The limited understanding of SM algorithms among the population suggest that adolescents are more likely to place blind trust in the information they consume, exposing them to negative consequences (misinformation, filter bubbles and echo chambers). We therefore propose an intervention with a narrative scripts approach to raise awareness of AI algorithms in SM. To foster an authentic learning experience and question adolescents’ trust in AI, we deploy a low-accuracy AI image classifier. A quasi-experimental study was conducted among 144 high-school students in Barcelona, Spain. The results show that the narrative scripts intervention improved students’ awareness of SM algorithms and shaped more critical attitudes towards them. A comparison of students’ choices between human predictions and those produced by a low-accuracy AI classifier shows a lack of AI overdependence. Information about predictions’ source did not affect students’ trust or learning about AI. These findings contribute towards SM algorithms education and share insight into the effect of deploying low-accuracy detectors in learning technology interventions.

The provision of toxic content and misinformation is a frequent phenomenon in current social media with specific impact and risks for younger users. We report on efforts taken in the project Courage to mitigate and overcome these threats through dedicated educational technology inspired by psychological and pedagogical approaches. The aim is to empower adolescents to confidently interact with and utilize social media and to increase their awareness and resilience. For this purpose, we have adopted approaches from the field of Intelligent Tutoring Systems, namely the provision of a virtual learning companion (VLC). The technical system is a browser-based environment that allows for combining a controllable social media space with a VLC as a plugin. This environment is backed by an API that bundles Machine Learning and Natural Language Processing algorithms for detecting and classifying different types of risks. The pedagogical scenarios that are supported by this technical environment and approach range from chat-based dialogues to more complex narrative scripts.

Fake news detection has become a research area that goes way beyond a purely academic interest as it has direct implications on our society as a whole. Recent advances have primarily focused on textbased approaches. However, it has become clear that to be effective one needs to incorporate additional, contextual information such as spreading behaviour of news articles and user interaction patterns on social media. We propose to construct heterogeneous social context graphs around news articles and reformulate the problem as a graph classification task. Exploring the incorporation of different types of information (to get an idea as to what level of social context is most effective) and using different graph neural network architectures indicates that this approach is highly effective with robust results on a common benchmark dataset.

Artificial intelligence's (AI) progress holds great promise in tackling pressing societal concerns such as health and climate. Large Language Models (LLM) and the derived chatbots, like ChatGPT, have highly improved the natural language processing capabilities of AI systems allowing them to process an unprecedented amount of unstructured data. However, the ensuing excitement has led to negative sentiments, even as AI methods demonstrate remarkable contributions (e.g. in health and genetics). A key factor contributing to this sentiment is the misleading perception that LLMs can effortlessly provide solutions across domains, ignoring their limitations such as hallucinations and reasoning constraints. Acknowledging AI fallibility is crucial to address the impact of dogmatic overconfidence in possibly erroneous suggestions generated by LLMs. At the same time, it can reduce fear and other negative attitudes toward AI. This necessitates comprehensive AI literacy interventions that educate the public about LLM constraints and effective usage techniques, i.e prompting strategies. With this aim, a pilot educational intervention was performed in a high school with 21 students. It involved presenting high-level concepts about intelligence, AI, and LLMs, followed by practical exercises involving ChatGPT in creating natural educational conversations and applying established prompting strategies. Encouraging preliminary results emerged, including high appreciation of the activity, improved interaction quality with the LLM, reduced negative AI sentiments, and a better grasp of limitations, specifically unreliability, limited understanding of commands leading to unsatisfactory responses, and limited presentation flexibility. Our aim is to explore AI acceptance factors and refine this approach for more controlled future studies.

In this report, we present the project URWalking conducted at the University of Regensburg. We describe its major outcomes: Firstly, an indoor navigation system for pedestrians as a web application and as an Android app with position tracking of users in indoor and outdoor environments. Our implementation showcases that a variant of the $$A^*$$ -algorithm by Ullmann (tengetriebene optimierung präferenzadaptiver fußwegrouten durch gebäudekomplexe https://epub.uni-regensburg.de/43697/ , 2020) can handle the routing problem in large, levelled indoor environments efficiently. Secondly, the apps have been used in several studies for a deeper understanding of human wayfinding. We collected eye tracking and synchronized video data, think aloud protocols, and log data of users interacting with the apps. We applied state-of-the-art deep learning models for gaze tracking and automatic classification of landmarks. Our results indicate that even the most recent version of the YOLO image classifier by Redmon and Farhadi (olov3: An incremental improvement. arXiv, 2018) needs finetuning to recognize everyday objects in indoor environments. Furthermore, we provide empirical evidence that appropriate machine learning models are helpful to bridge behavioural data from users during wayfinding and conceptual models for the salience of objects and landmarks. However, simplistic models are insufficient to reasonably explain wayfinding behaviour in real time—an open issue in GeoAI. We conclude that the GeoAI community should collect more naturalistic log data of wayfinding activities in order to build efficient machine learning models capable of predicting user reactions to routing instructions and of explaining how humans integrate stimuli from the environment as essential information into routing instructions while solving wayfinding tasks. Such models form the basis for real-time wayfinding assistance.

We present a digital media literacy activity composed of (i) an educational talk and (ii) a game-based activity. The aim is to support teachers in developing learning activities to increase awareness of social media threats among students. Through this activity students directly experience phenomena like echo chambers and filter bubbles that can be provoked by harmful online interaction dynamics controlled by social media platforms’ recommender systems while remaining invisible to the affected users. Our preliminary findings show that a game-based direct experience, inspired by the wisdom of crowds phenomenon, can increase the perception of social media influence on participants with statistically significant results compared to standard lecture-based activity. We conclude that developing a tool enabling educators and scholars to easily perform the proposed activity can be helpful to improve digital media literacy effectiveness.

Social media have become an integral part of our lives, expanding our interlinking capabilities to new levels. There is plenty to be said about their positive effects. On the other hand, however, some serious negative implications of social media have been repeatedly highlighted in recent years, pointing at various threats to society and its more vulnerable members, such as teenagers, in particular, ranging from much-discussed problems such as digital addiction and polarization to manipulative influences of algorithms and further to more teenager-specific issues (e.g., body stereotyping). The impact of social media—both at an individual and societal level—is characterized by the complex interplay between the users' interactions and the intelligent components of the platform. Thus, users' understanding of social media mechanisms plays a determinant role. We thus propose a theoretical framework based on an adaptive “Social Media Virtual Companion” for educating and supporting an entire community, teenage students, to interact in social media environments in order to achieve desirable conditions, defined in terms of a community-specific and participatory designed measure of Collective Well-Being (CWB). This Companion combines automatic processing with expert intervention and guidance. The virtual Companion will be powered by a Recommender System (CWB-RS) that will optimize a CWB metric instead of engagement or platform profit, which currently largely drives recommender systems thereby disregarding any societal collateral effect. CWB-RS will optimize CWB both in the short term by balancing the level of social media threats the users are exposed to, and in the long term by adopting an Intelligent Tutor System role and enabling adaptive and personalized sequencing of playful learning activities. We put an emphasis on experts and educators in the educationally managed social media community of the Companion. They play five key roles: (a) use the Companion in classroom-based educational activities; (b) guide the definition of the CWB; (c) provide a hierarchical structure of learning strategies, objectives and activities that will support and contain the adaptive sequencing algorithms of the CWB-RS based on hierarchical reinforcement learning; (d) act as moderators of direct conflicts between the members of the community; and, finally, (e) monitor and address ethical and educational issues that are beyond the intelligent agent's competence and control. This framework offers a possible approach to understanding how to design social media systems and embedded educational interventions that favor a more healthy and positive society. Preliminary results on the performance of the Companion's components and studies of the educational and psychological underlying principles are presented.

Educational chatbots come with a promise of interactive and personalized learning experiences, yet their development has been limited by the restricted free interaction capabilities of available platforms and the difficulty of encoding knowledge in a suitable format. Recent advances in language learning models with zero-shot learning capabilities, such as ChatGPT, suggest a new possibility for developing educational chatbots using a prompt-based approach. We present a case study with a simple system that enables mixed-turn chatbot interactions and discuss the insights and preliminary guidelines obtained from initial tests. We examine ChatGPT's ability to pursue multiple interconnected learning objectives, adapt the educational activity to users' characteristics, such as culture, age, and level of education, and its ability to use diverse educational strategies and conversational styles. Although the results are encouraging, challenges are posed by the limited history maintained for the conversation and the highly structured form of responses by ChatGPT, as well as their variability, which can lead to an unexpected switch of the chatbot's role from a teacher to a therapist. We provide some initial guidelines to address these issues and to facilitate the development of effective educational chatbots.

Natural language processing and other areas of artificial intelligence have seen staggering progress in recent years, yet much of this is reported with reference to somewhat limited benchmark datasets. We see the deployment of these techniques in realistic use cases as the next step in this development. In particular, much progress is still needed in educational settings, which can strongly improve users’ safety on social media. We present our efforts to develop multi-modal machine learning algorithms to be integrated into a social media companion aimed at supporting and educating users in dealing with fake news and other social media threats. Inside the companion environment, such algorithms can automatically assess and enable users to contextualize different aspects of their social media experience. They can estimate and display different characteristics of content in supported users’ feeds, such as ‘fakeness’ and ‘sentiment’, and suggest related alternatives to enrich users’ perspectives. In addition, they can evaluate the opinions, attitudes, and neighbourhoods of the users and of those appearing in their feeds. The aim of the latter process is to raise users’ awareness and resilience to filter bubbles and echo chambers, which are almost unnoticeable and rarely understood phenomena that may affect users’ information intake unconsciously and are unexpectedly widespread. The social media environment is rapidly changing and complex. While our algorithms show state-of-the-art performance, they rely on task-specific datasets, and their reliability may decrease over time and be limited against novel threats. The negative impact of these limits may be exasperated by users’ over-reliance on algorithmic tools. Therefore, companion algorithms and educational activities are meant to increase users’ awareness of social media threats while exposing the limits of such algorithms. This will also provide an educational example of the limits affecting the machine-learning components of social media platforms. We aim to devise, implement and test the impact of the companion and connected educational activities in acquiring and supporting conscientious and autonomous social media usage.

Recent progress in natural language processing has been impressive in many different areas with transformer-based approaches setting new benchmarks for a wide range of applications. This development has also lowered the barriers for people outside the NLP community to tap into the tools and resources applied to a variety of domain-specific applications. The bottleneck however still remains the lack of annotated gold-standard collections as soon as one’s research or professional interest falls outside the scope of what is readily available. One such area is genocide-related research (also including the work of experts who have a professional interest in accessing, exploring and searching large-scale document collections on the topic, such as lawyers). We present GTC (Genocide Transcript Corpus), the first annotated corpus of genocide-related court transcripts which serves three purposes: (1) to provide a first reference corpus for the community, (2) to establish benchmark performances (using state-of-the-art transformer-based approaches) for the new classification task of paragraph identification of violence-related witness statements, (3) to explore first steps towards transfer learning within the domain. We consider our contribution to be addressing in particular this year’s hot topic on Language Technology for All.

In this paper we propose a deep learning model based on graph machine learning (i.e. Graph Attention Convolution) and a pretrained transformer language model (i.e. ELECTRA). Our model was developed to detect harmful tweets about COVID-19 and was used to tackle subtask 1C (harmful tweet detection) at the CheckThat!Lab shared task organized as part of CLEF 2022. In this binary classification task, our proposed model reaches a binary F1 score (positive class label, i.e. harmful tweet) of 0.28 on the test set. We demonstrate that our approach outperforms the official baseline by 8% and describe our model as well as the experimental setup and results in detail. We also refer to limitations of the approach and future research directions. © 2022 Copyright (c) 2022 for this paper by its authors.

The rise of deep learning methods has transformed the research area of natural language processing beyond recognition. New benchmark performances are reported on a daily basis ranging from machine translation to question-answering. Yet, some of the unsolved practical research questions are not in the spotlight and this includes, for example, issues arising at the interface between spoken and written language processing. We identify sentence boundary detection and speaker change detection applied to automatically transcribed texts as two NLP problems that have not yet received much attention but are nevertheless of practical relevance. We frame both problems as binary tagging tasks that can be addressed by fine-tuning a transformer model and we report promising results.

In this demo, we present an interactive recommender system that suggests recipes to participants of a weight loss programme. Nutritional constraints imposed by the programme serve as initial information to tackle the crucial cold start problem.

This paper describes our approach (URmSBD) to address the shared task on Sentence End and Punctuation Prediction in NLG Text (SEPP-NLG) organised as part of SwissText 2021. We participated in Subtask 1 (fully unpunctuated sentences – full stop detection) and submitted a run for every featured language (English, German, French, Italian). Our submissions are based on pre-trained BERT models that have been fine-tuned to the task at hand. We had recently demonstrated, that such an approach achieves state-of-the-art performance when identifying end-of-sentence markers on automatically transcribed texts. The difference to that work is that here we use languagespecific BERT models for each featured language. By framing the problem as a binary tagging task using the outlined architecture we are able to achieve competitive results on the official test set across all languages, with Recall, Precision, F1 ranging between 0.91 and 0.96 which makes us joint winners for Recall in two of the languages. The official baselines are beaten by large margins.

We identify automated landmark salience assessment in indoor environments as a problem related to pedestrian navigation systems that has not yet received much attention but is nevertheless of practical relevance. We therefore evaluate an approach based on visual information using images to capture the landmarks’ outward appearance. In this context we introduce the largest landmark image and salience value data set in the domain so far. We train various classifiers on domain agnostic visual features to predict the salience of landmarks. As a result, we are able to clarify the role of visual object features regarding perception of landmarks. Our results demonstrate that visual information has only limited expressiveness with respect to salience.