Marina Delianidi

This paper introduces DK-PRACTICE, an intelligent educational platform that combines Knowledge Tracing (KT) and recommendation systems to support personalized learning in higher education. The platform utilizes a novel Paired Bipolar Bag-of-Words (PB-BoW) model to assess students’ knowledge states, forecast performance, and offer targeted recommendations. To test its effectiveness in real-world settings, DK-PRACTICE was implemented in the “Computer Organization and Architecture” undergraduate course, involving 138 students in Pre-Test and 106 in Post-Test. Empirical analysis of benchmark datasets and a newly created course dataset showed that the PB-BoW model outperformed an RNN-based KT model in predictive accuracy. Student surveys indicated high levels of satisfaction with usability, relevance of recommendations, and overall learning support, with most participants expressing willingness to reuse the platform in other courses. These results demonstrate the potential of DK-PRACTICE as a scalable and adaptable tool for improving personalized learning and bridging the gap between AI-driven KT research and classroom implementation.

This study presents a production-ready AI pipeline for audio content processing, implemented within the Youth Radio platform, which serves as an extension of the European School Radio initiative. The system uses a multi-server architecture: an AI Server that runs batch/offline jobs, orchestrated by Apache Airflow, and two Web Servers that deliver all the Backend as well as the Frontend applications, configured with load balancing and redundancy to ensure high availability and fault tolerance. The implemented AI Pipeline includes tasks such as preprocessing, transcription, audio classification and topic modeling. Processed Podcasts are indexed in a Qdrant vector database to facilitate both dense and sparse retrieval while a recommendation system enriches the user’s experience. We summarize design choices and report system-level metrics and task-level indicators (ASR quality after correction, retrieval effectiveness) to guide similar deployments.

This paper introduces DK-PRACTICE, an intelligent educational platform that combines Knowledge Tracing (KT) and recommendation systems to support personalized learning in higher education. The platform utilizes a novel Paired-Bipolar Bag-of-Words (PB-BoW) model to assess students’ knowledge states, forecast performance, and offer targeted recommendations. To test its effectiveness in real-world settings, DK-PRACTICE was implemented in the “Computer Organization and Architecture” undergraduate course, involving 138 students in Pre-Test and 106 in Post-Test. Empirical analysis of benchmark datasets and a newly created course dataset showed that the PB-BoW model outperformed an RNN-based KT model in predictive accuracy. Student surveys indicated high levels of satisfaction with usability, relevance of recommendations, and overall learning support, with most participants expressing willingness to reuse the platform in other courses. These results demonstrate the potential of DK-PRACTICE as a scalable and adaptable tool for improving personalized learning and bridging the gap between AI-driven KT research and classroom implementation.

This study investigates the potential of language models to improve the classification of labor market information by linking job vacancy texts to two major European frameworks: the European Skills, Competences, Qualifications and Occupations (ESCO) taxonomy and the European Qualifications Framework (EQF). We examine and compare two prominent methodologies from the literature: Sentence Linking and Entity Linking. In support of ongoing research, we release an open-source tool, incorporating these two methodologies, designed to facilitate further work on labor classification and employment discourse. To move beyond surface-level skill extraction, we introduce two annotated datasets specifically aimed at evaluating how occupations and qualifications are represented within job vacancy texts. Additionally, we examine different ways to utilize generative large language models for this task. Our findings contribute to advancing the state of the art in job entity extraction and offer computational infrastructure for examining work, skills, and labor market narratives in a digitally mediated economy. Our code is made publicly available: https://github.com/tabiya-tech/tabiya-livelihoods-classifier

This study presents a production-ready AI pipeline for audio content processing, implemented within the Youth Radio platform (www.youthradio.eu), which serves as an extension of the European School Radio initiative (www.europeanschoolradio.eu). The system uses a multi-server architecture: an AI Server that runs batch/offline jobs, orchestrated by Apache Airflow, and two Web Servers that deliver all the Backend as well as the Frontend applications, configured with load balancing and redundancy to ensure high availability and fault tolerance. The implemented AI-Pipeline includes tasks such as preprocessing, transcription, audio classification and topic modelling. Processed Podcasts are indexed in a Qdrant vector database to facilitate both dense and sparse retrieval while a recommendation system enriches the user’s experience. We summarize design choices and report system-level metrics and task-level indicators (ASR quality after correction, retrieval effectiveness) to guide similar deployments.

Research into session-based recommendation systems (SBSR) has attracted a lot of attention, but each study focuses on a specific class of methods. This work examines and evaluates a large range of methods, from simpler statistical co-occurrence methods to embeddings and SotA deep learning methods. This paper analyzes theoretical and practical issues in developing and evaluating methods for SBSR in e-commerce applications, where user profiles and purchase data do not exist. The major tasks of SBRS are reviewed and studied, namely: prediction of next-item, next-basket and purchase intent. For physical retail shopping where no information about the current session exists, we treat the previous baskets purchased by the user as previous sessions drawn from a loyalty system. Mobile application scenarios such as push notifications and calling tune recommendations are also presented. Recommender models using graphs, embeddings and deep learning methods are studied and evaluated in all SBRS tasks using different datasets. Our work contributes a number of very interesting findings. Among all tested models, LSTMs consistently outperform other methods of SBRS in all tasks. They can be applied directly because they do not need significant fine-tuning. Additionally, they naturally model the dynamic browsing that happens in e-commerce web applications. On the other hand, another important finding of our work is that graph-based methods can be a good compromise between effectiveness and efficiency. Another important conclusion is that a “temporal locality principle” holds, implying that more recent behavior is better suited for prediction. In order to evaluate these systems further in realistic environments, several session-based recommender methods were integrated into an e-shop and an A/B testing method was applied. The results of this A/B testing are in line with the experimental results, which represents another important contribution of this paper. Finally, important parameters such as efficiency, application of business rules, re-ranking issues, and the utilization of hybrid methods are also considered and tested, providing comprehensive useful insights into SBRS and facilitating the transferability of this research work to other domains and recommendation scenarios.

Conventional recommendation methods such as collaborative filtering cannot be applied when long-term user models are not available. In this paper, we propose two session-based recommendation methods for anonymous browsing in a generic e-commerce framework. We represent the data using a graph where items are connected to sessions and to each other based on the order of appearance or their co-occurrence. In the first approach, called Hierarchical Sequence Probability (HSP), recommendations are produced using the probabilities of items’ appearances on certain structures in the graph. Specifically, given a current item during a session, to create a list of recommended next items, we first compute the probabilities of all possible sequential triplets ending in each candidate’s next item, then of all candidate item pairs, and finally of the proposed item. In our second method, called Recurrent Item Co-occurrence (RIC), we generate the recommendation list based on a weighted score produced by a linear recurrent mechanism using the co-occurrence probabilities between the current item and all items. We compared our approaches with three state-of-the-art Graph Neural Network (GNN) models using four session-based datasets one of which contains data collected by us from a leather apparel e-shop. In terms of recommendation effectiveness, our methods compete favorably on a number of datasets while the time to generate the graph and produce the recommendations is significantly lower.

The work in this paper is an extended research of our previous work: M. Delianidi, K. Diamantaras, G. Chrysogonidis, and V. Nikiforidis,“Student performance prediction using dynamic neural models,” in Fourteenth International Conference on Educational Data Mining (EDM 2021), 2021, pp. 46–54. In both works we study the task of predicting a student’s performance in a series of questions based, at each step, on the answers he/she has given on the previous questions. We propose a recurrent neural network approach where the dynamic part of the model is a Bidirectional GRU layer. In this work, we differentiate the model architecture from the earlier paper by imposing that the dynamic part is based exclusively on the history of previous question/answers, not including the current question. Then, the subsequent classification part is fed by the output of the dynamic part and the current question. In this way, the first part estimates the student’s knowledge state and represents it with a dynamically generated vector considering only the student’s previous questions and responses. We call this part the “Knowledge State Representation subnet’‘. Using this representation, the following “Tracing subnet’‘ which is a static multi-layer classifier can predict the correctness of the answer to any following question. Therefore, this architecture is suitable not only for described prediction task but also for recommendation tasks which were not possible with the previous architecture. In addition, in this work, we also study the effectiveness of our models against each other, as well as against the previous state-of-the-art models on seven datasets including a new multi-skills dataset derived from the NeurIPS 2020 education challenge. Both our models compare favorably against the state-of-the-art methods in almost all datasets, while the newly proposed architecture achieves very similar results compared to our earlier model in most cases except for two cases, including the multi-skill “NeurIPS-2020-small” dataset, where it achieves considerably better results.