Graph Neural Networks (GNNs) have demonstrated significant success in learning from graph-structured data but often struggle on heterophilous graphs, where connected nodes differ in features or class labels. This limitation arises from indiscriminate neighbor aggregation and insufficient incorporation of higher-order structural patterns. To address these challenges, we propose GLANCE (Graph Logic Attention Network with Cluster Enhancement), a novel framework that integrates logic-guided reasoning, dynamic graph refinement, and adaptive clustering to enhance graph representation learning. GLANCE combines a logic layer for interpretable and structured embeddings, multi-head attention-based edge pruning for denoising graph structures, and clustering mechanisms for capturing global patterns. Experimental results in benchmark datasets, including Cornell, Texas, and Wisconsin, demonstrate that GLANCE achieves competitive performance, offering robust and interpretable solutions for heterophilous graph scenarios. The proposed framework is lightweight, adaptable, and uniquely suited to the challenges of heterophilous graphs.

Accurate prediction of gas concentrations at longwall mining faces is critical for safety production, yet current methods still face challenges in interpretability and reliability. This study aims to enhance prediction accuracy and model interpretability by employing advanced feature selection techniques. We integrate Shapley Additive Explanations (SHAP) into feature selection process to identify and quantify the contributions of multivariate features to gas concentration variations. The effectiveness of SHAP-based feature selection is systematically evaluated alongside Principal Component Analysis, Dynamic Time Warping, and unfiltered features, across four baseline predictive models chosen based on their structural characteristics: Long Short-Term Memory, Gated Recurrent Unit, Transformer and Graph Neural Network. Using public dataset from the Upper Silesian coal basin in Poland, we demonstrate that models trained with SHAP-selected features outperform baseline models, particularly in terms of accuracy and reliability for long-term predictions. By identifying the most relevant features and clarifying their interactions, this study enhances predictive performance and provides deeper insights into the dynamics governing gas concentrations, emphasising the value of advanced, interpretable feature selection techniques in developing robust models for industrial applications in mining.

The global proliferation of Massive Open Online Courses (MOOCs) has led to an increasing significance of personalised course recommenders. While existing recommendation approaches that combine Graph Neural Networks (GNNs) and knowledge graphs have been effective in enriching feature representations and addressing the personalised needs of learners, GNNs still encounter significant challenges, global information loss caused by long-range dependencies issue when processing large-scale graph data. To address these limitations, this paper proposes, for the first time, a novel MOOC recommendation model that integrates knowledge graph, GNNs, positional encoding, and contrastive learning, leveraging graph patches and embedding patches. By partitioning large-scale knowledge graph data into graph patches and decomposing embeddings, along-side the incorporation of contrastive learning techniques, the proposed model effectively extracts the local and global node information from different views. Experimental evaluations conducted on two real-life educational datasets (COCO and Xuetang) demonstrate the effectiveness of the proposed model in improving recommendation accuracy and enhancing performance.

Expressing information in different forms is an important skill that students should develop in school. This skill positively impacts academic reading and writing. However, it can also lead to negative consequences, such as plagiarism. Students may paraphrase original texts and present them as their own work. Therefore, the need to develop effective approaches to detect plagiarism and identify paraphrase has become increasingly important in academia, journalism, publishing, and other fields where innovation, novelty, and originality are highly valued, especially with the rising incidence of plagiarism in these areas because of the easy access to information on the internet and the capabilities of large language models. Most published detection methods analyse plagiarism at the sentence-level. We have developed approaches for generating and detecting paraphrased paragraphs by considering inter-sentence and intra-sentence relations, which enables the identification of paraphrased text at the paragraph-level. This includes joining, splitting, and/or shifting sentences within a paragraph, as students often plagiarise paragraphs. In the generating stage, we create the ALECS dataset, by developing three algorithms and applying a masking approach to tackle the paragraph’s syntactic and lexical layers while maintaining the paragraph’s semantics. ALECS can contribute to developing students’ abilities in paraphrasing, as there are more than 6 different forms for each source paragraph. In addition, as in this study, ALECS can be employed to train deep learning models for the purpose of generating or detecting plagiarised paragraphs. For the detection phase, our method shows robust results and outperforms existing work in detecting paragraph-level paraphrases, achieving a 90.1 F1 score with Longformer and reaching 96 when using a fine-tuned GPT-3.5.

Massive Open Online Courses (MOOCs) have been proliferating, offering free or low-cost content for learners. Nevertheless, the certification rate of both free and paid courses has been low (between 4.5% - 13% and 1% - 3%, respectively). Thus, this study aims to survey MOOCs certification predictive models, synthesise results for a comprehensive and deep understanding of this field and explore how these models contributed to addressing the very low certification level. We adopted the Preferred Reporting Items for Systematic Review and Meta-analysis Protocols (PRISMA-P) for transparently conducting the present review and reporting the results from the works reviewed. Additionally, this SLR highlights several trends and limitations within the present predictive models, including some methodological concerns: the extent to which the present models are generalisable, the excessive filtration of the experimental population, the incompatibility of some experiments with real-time scenarios (nonrealistic modelling), and the shallow reporting of model performances. We have also discussed the replicability of the present models and ongoing efforts towards building a state-of-the-art predictive model. Finally, we highlight future research opportunities in the field of MOOC certification prediction that either deal with the limitations of the present models or address unanswered questions.

The generation and identification of paraphrases are essential challenges in natural language processing (NLP), involving considerable consequences for education, plagiarism detection, and content analysis. Despite significant advancements in sentence-level paraphrasing, paragraph-level paraphrasing is still inadequately investigated, especially across various fields. This research is the first to examine the effectiveness of SALAC algorithms and Transformer-based models in generating coherent and semantically accurate paraphrases across various domains, utilising the ALECS dataset, which includes text samples from five different educational fields, such as Economics and Anthropology. The methodology uses SALAC algorithms for sentence reordering, according to coherence scores obtained from the ALBERT model's sentence Order Prediction (SOP). Real-world human evaluation is performed on the paraphrased paragraphs, measuring semantic similarity and coherence via a Likert scale of five pre-defined points. Evaluators involve academic students and researchers, ensuring the reliability of outcomes. Research findings show that SALAC algorithms successfully maintain semantic integrity and coherence across several domains, highlighting their generalisability. Additionally, the study examines how domain-specific readability impacts annotation reliability and paraphrasing performance. This research contributes to the development of robust, domain-independent paraphrasing techniques at the paragraph-level for educational applications and plagiarism detection, thereby advancing NLP solutions across multiple domains.

Most knowledge tracing systems prioritise performance prediction over the delivery of personalised, actionable feedback. This paper proposes SPAR-GNN (Selective Pattern-Aware Reasoning with Graph Neural Networks), a novel framework that combines heterogeneous GNN-based modelling with selectively triggered feedback from a Large Language Model (LLM). SPAR-GNN represents students, problems, skills and virtual pattern nodes that encode latent behavioural traits such as frustration and hint overuse within a unified graph structure. A customised Heterogeneous Graph Layer learns student representations, while the LLM is activated only for at-risk learners identified through behavioural and performance indicators. This reduces computational overhead while enhancing pedagogical relevance. SPAR-GNN consistently outperforms strong baselines across multiple metrics, demonstrating both predictive accuracy and robustness. Observational analysis also reveals interpretable links between behaviours and learning progression, supporting fairness-aware interventions.

The integration of AI into education enables more flexible and effective learning. Large Language Models (LLMs) such as ChatGPT offer broad topic coverage but lack personalization and may generate irrelevant or inaccurate content. To address these challenges, we propose TutorLLM, a personalized learning system that combines Knowledge Tracing (KT) and Retrieval-Augmented Generation (RAG). TutorLLM tailors responses based on each student’s learning state, predicted by the MLFBK KT model, and improves relevance using a Scraper component for context retrieval. Implemented as a Chrome plugin, TutorLLM was evaluated in a two-week field study with undergraduate students, demonstrating a 10

Cluster-based topic modelling has been demonstrated to be an effective method for identifying underlying topics within a corpus of text. Such techniques present the opportunity to leverage powerful embedding models to encode textual semantics in embeddings, before identifying dense clusters of these embeddings as representing topics. However, a key necessity for this process is a prior reduction in the dimensionality of high-dimensional embeddings to ensure efficient clustering. Currently, the UMAP algorithm represents the state-of-art algorithm used in topic modelling algorithms such as Top2Vec for this task. In this work, we investigate a novel paradigm in topic modelling, facilitated by the parametric UMAP algorithm. We propose to investigate how the architectural design of neural networks can contribute to parametric dimesionality reduction, to ensure high-quality topic modelling solutions. We achieve this by implementing a modified transformer-encoder architecture, with novel additional residual connections, into a dimensionality reduction pipeline in the benchmark cluster-based topic model Top2Vec, demonstrating the effectiveness of the addition through an in-depth topic analysis from both a metric and quality perspective. The analysis indicates that incorporating the transformer-encoder architecture for parametric dimensionality reduction in Top2Vec results in an enhancement, as measured by widely accepted topic evaluation metrics, which is further enhanced by the introduction of additional residual connections into the network architecture. Moreover, upon human assessment of the identified topics, it is evident that the proposed transformer-encoder pipelines enhance the granularity of the topic modelling solution, when dealing with a small dataset.

As the business world shifts to the web and tremendous amounts of data become available on multilingual mobile applications, new business and research challenges and opportunities have been explored. This research aims to intensify the usage of data analytics, machine learning, and sentiment analysis of textual data to classify customers’ reviews, feedback, and ratings of businesses in Jordan’s food and restaurant industry. The main methods used in this research were sentiment polarity (to address the challenges posed by businesses to automatically apply text analysis) and bio-metric techniques (to systematically identify users’ emotional states, so reviews can be thoroughly understood). The research was extended to deal with reviews in Arabic, dialectic Arabic, and English, with the main focus on the Arabic language, as the application examined (Talabat) is based in Jordan. Arabic and English reviews were collected from the application, and a new model was proposed to sentimentally analyze reviews. The proposed model has four main stages: data collection, data preparation, model building, and model evaluation. The main purpose of this research is to study the problem expressed above using a model of ordinal regression to overcome issues related to misclassification. Additionally, an automatic multi-language prediction approach for online restaurant reviews was proposed by combining the eXtreme gradient boosting (XGBoost) and particle swarm optimization (PSO) techniques for the ordinal regression of these reviews. The proposed PSO-XGB algorithm showed superior results when compared to support vector machine (SVM) and other optimization methods in terms of root mean square error (RMSE) for the English and Arabic datasets. Specifically, for the Arabic dataset, PSO-XGB achieved an RMSE value of 0.7722, whereas PSO-SVM achieved an RSME value of 0.9988.

Abstract reasoning is a key ability for students’ cognitive development, yet traditional methods often fail to provide an accurate and objective assessment. We propose a novel machine learning approach that utilises EEG data from low-cost headsets to predict response correctness in abstract reasoning tasks. This paper presents an adaptive LSTM model incorporating multi-head attention for analysing EEG data acquired during both the reasoning (pre-response) and feedback (post-response) phases of abstract reasoning questions. Results demonstrate the model’s ability to predict answer correctness with promising performance across multiple evaluation metrics. Additionally, we investigate the influence of self-reported confidence on EEG-based correctness prediction, yielding valuable insights. Our model demonstrates scalability and robustness when applied to larger datasets. This research paves a way for prospective applications of innovative tools in personalised feedback and adaptive learning systems, and advances methods for assessing and supporting cognitive growth.

Students’ success in the 21st century demands not only strong academic skills but also well-developed Non-academic Skills and Abilities (NaSAs) such as critical thinking, concentration, and emotion regulation. The emerging field of Brain-Computer Interfaces (BCIs) offers intriguing possibilities for enhancing student learning and development, by providing real-time neurofeedback that could inform personalised strategies. While research on the impact of BCIs on academic skills is growing, there is limited research regarding their potential to promote NaSAs. This Systematic Literature Review (SLR) aims to address this gap, by investigating and analysing the existing research on how BCIs can be used to assist the development of these crucial skills and abilities. This SLR provides a comprehensive analysis of 46 empirical studies between 2013–2023 selected via the well-accepted PRISMA method from 922 candidate studies. This analysis explores the diverse ways BCIs can facilitate or enhance the development of cognitive, motor, and inter-/intra- personal NaSAs, either passively or actively. The findings of this SLR offer valuable insights into the potential of BCIs to revolutionise education towards a neuro-enhanced future that promotes not only academic achievement, but also holistic student growth .

Gamification applied to learning environments is widely accepted as positively impacting students' psychological and cognitive aspects, such as motivation and learning performance. According to the literature on the subject, gamification tends to promote more positive effects on students than negative ones. Meanwhile, the literature lacks a deeper understanding of how education professionals perceive gamification in learning environments and their concerns about implicit issues and ethical issues. Prior research has not examined the relationship between gamification in education, its ethical concerns, and barriers. As a result, we expanded a previous study to identify and delve deep into potential barriers and ethical concerns pertaining to gamification from the perspective of Brazilian teachers. A survey was designed and answered by 61 Brazilian teachers. According to our findings, teachers are not inclined to use gamification for various reasons, such as social factors (e.g., acceptance by teachers and students) and planning and evaluation issues (e.g., lack of knowledge). Our study also found that their ethical concerns pertain to psychological effects, social issues, privacy issues, humanization, and behavioral effects. As part of the contribution of this paper, we list potential barriers and ethical concerns that designers and researchers should keep in mind when designing and implementing gamification and gamification-based personalization in learning environments.

EDITORIAL article Front. Hum. Neurosci., 08 March 2024Sec. Cognitive Neuroscience Volume 18 - 2024 | https://doi.org/10.3389/fnhum.2024.1390788

The topic of immersive virtual reality (IVR) in education has gained increasing attention in recent years, due to its potential to enhance learner outcomes and to mitigate learning costs. As we can capture a multitude of information from speech and generate valuable information from it, there has been an interest in exploring this source of data in such environments. Additionally, speech is being used in new and different ways in such environments. However, its specific usage in IVR-based education has not been reviewed yet. Thus, this systematic review seeks to examine, for the first time, the current state of research, specifically on using speech input - and, related to this, Natural Language Processing (NLP) - in educational IVR environments. We conducted a comprehensive search of the popular Web of Science and Scopus databases, to identify relevant papers. To properly reflect the state-of-the-art, English peer-reviewed articles published in the last 5 years (between 2020 - 2024), were included in the review, based on keywords search. 595 articles were identified and processed via the established PRISMA procedure, to exclude all duplicate or irrelevant papers, rendering 23 articles as relevant. For these, we identified the target educational subjects and the purpose of using speech as a data source. We also investigated speech recognition models and NLP models used. This systematic review provides evidence supporting the use of speech input as a valuable data source in educational IVR applications. We also propose the first, to the best of our knowledge, taxonomy for speech and NLP for IVR in education, as well as identify potential further research directions, all of which can help researchers and educators, when considering incorporating speech and NLP into educational IVR applications, to enrich the teaching and learning experience.

Massive Open Online Courses (MOOCs) have revolutionised the landscape of education, generating unprecedented arrays of diverse data about, from and by learners. However, there are few publicly accessible large-scale text datasets that several stakeholders, including course designers and instructors, can use. This paper presents MOOCRev, a groundbreaking data repository aggregating MOOC reviews from various platforms, leveraging cutting-edge data collection and processing techniques. The dataset encompasses many review characteristics (over 1,250,000 course reviews and ratings), making it, to the best of our knowledge, the largest publicly accessible MOOC review dataset. MOOCRev targets the following challenges (1) thorough, research-based anonymisation of the data to ensure user privacy and data integrity, (2) dealing with class (rating) bias, (3) addressing language diversity and platform-specific rating systems, (4) processing emojis and emoticons, (5) providing text weights (6) and tagging the reviews with the corresponding Parts of Speech (POS). The repository provides researchers, educators, and platform developers with invaluable opportunities for various Natural Language Processing (NLP) tasks, including learners' sentiments and feedback analysis, aiding future learners in selecting courses aligned with their interests and learning styles based on previous learners' experiences.

The availability and growth of tools and natural language generation (NLG) models that are used to paraphrase text could be helping to improve students' writing and comprehension skills or a threat to intellectual property and educational integrity specifically when the text has been copied from other authors. These tools can be used by plagiarists to paraphrase individual words, phrases, sentences, and paragraphs. To solve this issue, much work has been done on plagiarism detection (PD) and paraphrase identification (PI) utilising downstream tasks and natural language processing (NLP) methods. These works mainly focus on sentence length and sentence-level paraphrasing. In this paper, we investigate paragraph-length and paragraph-level paraphrasing as the most common method of committing plagiarism is copying and paraphrasing paragraphs from other authors. Here, we construct a novel, large-scale paragraph-level paraphrasing dataset by implementing and examining a state-of-the-art Transformer-based model to reorder and paraphrase sentences without affecting a paragraph's meaning. In a first-of-a-kind study, we consider both intra-sentence and inter-sentence similarity before examining the efficiency of state-of-the-art Transformer-based models in detecting paraphrased paragraphs. We offer a technique that serves as both a tool for honing paraphrasing skills and a means of identifying plagiarism. Our outcomes surpass those presented in the existing literature.

Sentiment analysis (SA) has been, and is still, a thriving research area. However, the task of Arabic sentiment analysis (ASA) is still underrepresented in the body of research. This study offers the first in-depth and in-breadth analysis of existing ASA studies of textual content and identifies their common themes, domains of application, methods, approaches, technologies and algorithms used. The in-depth study manually analyses 133 ASA papers published in the English language between 2002 and 2020 from four academic databases (SAGE, IEEE, Springer, WILEY) and from Google Scholar. The in-breadth study uses modern, automatic machine learning techniques, such as topic modelling and temporal analysis, on Open Access resources, to reinforce themes and trends identified by the prior study, on 2297 ASA publications between 2010-2020. The main findings show the different approaches used for ASA: machine learning, lexicon-based and hybrid approaches. Other findings include ASA 'winning' algorithms (SVM, NB, hybrid methods). Deep learning methods, such as LSTM can provide higher accuracy, but for ASA sometimes the corpora are not large enough to support them. Additionally, whilst there are some ASA corpora and lexicons, more are required. Specifically, Arabic tweets corpora and datasets are currently only moderately sized. Moreover, Arabic lexicons that have high coverage contain only Modern Standard Arabic (MSA) words, and those with Arabic dialects are quite small. Thus, new corpora need to be created. On the other hand, ASA tools are stringently lacking. There is a need to develop ASA tools that can be used in industry, as well as in academia, for Arabic text SA. Hence, our study offers insights into the challenges associated with ASA research and provides suggestions for ways to move the field forward such as lack of Dialectical Arabic resource, Arabic tweets, corpora and data sets for SA.

The advent of advanced AI systems, particularly generative pre-trained transformers (GPTs) and large language models (LLMs) like ChatGPT, Gemini, Claude, has significantly changed the landscape of artificial intelligence and its applications across various domains. GPTs have demonstrated impressive capabilities in generating human-like text and solving complex problems. Educators are finding creative ways to deploy GPTs systems in their practice [1]. Many schools are struggling with the question of whether to allow or ban the use of AI, because of the potential risks (overreliance, academic integrity). The anticipation of change in the education system is palpable. What does the future hold? This raises important questions about the role of artificial intelligence in education (AIED) research. This panel will discuss the reasons why and ways in which AIED research should or should not remain essential in these transformative times for education, as well as ways in which it should change.

Agent-Based Models (ABMs) have been used in the field of education as a learning tool. However, the use of ABM as a tool for educational stakeholders is not well-represented, nor machine learning (ML) for ABM. Here we extend our work on a classroom ABM, by developing an ABM & ML hybrid model that simulates classroom disruptive interactions during a school year, and outputs predicted learning outcomes. We use real-life data from a primary school monitoring system that contains 65,385 student records from 2,040 schools across the UK as well as simulated interaction data, to implement linear regression for predictions of math scores, and ABM interactions to update the final score, to reflect the effect of these interactions. We show that this hybrid ABM model outperforms a simple ABM model.