Συνέδρια ΔΙΠΑΕ

Μόνιμο URI για αυτήν την κοινότηταhttps://repository.ihu.gr/handle/11544/48408

Περιηγούμαι

Πλοήγηση Συνέδρια ΔΙΠΑΕ ανά Θέμα "AI in education"

A B C D E F G H I J K L M N O P Q R S T U V W X Y Z
Α Β Γ Δ Ε Ζ Η Θ Ι Κ Λ Μ Ν Ξ Ο Π Ρ Σ Τ Υ Φ Χ Ψ Ω
Φιλτράρετε τα αποτελέσματα πληκτρολογώντας τα πρώτα γράμματα
Τώρα δείχνει 1 - 2 από 2
  • Μικρογραφία εικόνας
    Τεκμήριο
    Artificial Intelligence-Enhanced Literature Teaching: A Didactic Intervention Based on C. P. Cavafy’s “Ithaca”
    (2025-12-12) Kolitsi, Filothei; Πολυτεχνική Σχολή, Τμήμα Μηχανικών Πληροφορικής και Ηλεκτρονικών Συστημάτων
    Abstract—This paper presents a fully developed, AI‑enhanced teaching scenario for upper secondary literature education centered on the poem “Ithaca”. The instructional design combines close reading, comparative interpretation, collaborative inquiry, and multimodal creative production supported by AI tools. The aim is to connect poetic meaning with students’ lived experience while cultivating critical, digital, and multimodal literacies. The scenario is structured as a two‑hour intervention implemented in a computer lab and based on a flipped classroom model. Students engage with the poem and three “Cavafyesque” texts prior to class and then participate in a sequence of guided in‑class activities including audiovisual stimulus, word‑cloud elicitation, expressive listening, personal response writing, comparative thematic and stylistic analysis, AI‑based visualization, AI‑assisted music generation, and reflective self‑assessment. AI tools are positioned as creative and analytical instruments rather than answer engines, and all generated outputs must be interpretively justified by learners. The teacher acts as facilitator and coordinator, supporting dialogue, collaboration, and critical reflection. The design also incorporates comparison between human‑authored and AI‑generated literary texts in order to strengthen students’ awareness of style, authorship, and textual quality. The paper details the pedagogical framework, learning objectives, instructional phases, assessment strategy, and classroom management model. It argues that AI‑supported multimodal pedagogy can deepen poetry engagement when grounded in interpretive practice, collaborative meaning‑making, and reflective evaluation.
  • Μικρογραφία εικόνας
    Τεκμήριο
    Evaluation of Conversational Artificial Intelligence Agents for Automated Grading in Education
    (2025-12-12) Keroglou, Christoforos; Diamantaras, Konstantinos; Πολυτεχνική Σχολή, Τμήμα Μηχανικών Πληροφορικής και Ηλεκτρονικών Συστημάτων
    Abstract – This work focuses on evaluating the performance of three popular AI conversational agents (Gemini 2.5 Pro, ChatGPT 4.1, and Claude 3.7 Sonnet) as automated grading systems on a dataset containing short answers to questions from the subject area of Data Structures (Computer Science). The performance analysis was based on four main research questions: 1) How accurate is the evaluation of an AI-based automated grading system compared to the grading performed by expert human evaluators, 2) Is there a correlation between the length of the response (number of words) and the performance of AI-based automated grading systems, 3) Is there a correlation between the score of the response (as determined by expert human evaluators) and the performance of AI-based automated grading systems, 4) How confident can we be in the results provided by each system. Specifically, what variation exists in the outcomes when we repeat our experiments. Through these research questions, the study aims to provide a clearer understanding of the advantages and limitations of the three systems in their application within the educational system. The survey results show that ChatGPT 4.1 had the highest accuracy in automatic scoring, with a lower Root Mean Square Error (RMSE) compared to Claude 3.7 Sonnet and Gemini 2.5 Pro. Specifically, ChatGPT showed an RMSE of 0.82, followed by Claude with 1.02 and finally Gemini with 1.07. The statistical analysis for the whole dataset shows that the three agents had a relatively high and almost identical correlation coefficient with the grade of the expert raters, revealing the possibility of imitating human behavior in terms of scoring and at the same time indicating perhaps a common factor in the scoring process. However, there were significant differences in the performance of the three agents depending on the type of question. These differences suggest that each of the systems has distinct strengths and weaknesses depending on the content and structure of the question. Furthermore, extended research is needed to prove/or disprove the correlation (if any) between the number of words/or the score of the response (by expert human evaluators) with the error produced by the automated systems. The results of the present work are a first indication that each of the three AI systems may be suitable for use as an automatic grading system, which could be exploited in the future to ensure maximum accuracy in examination procedures or to simplify procedures involving courses with many students (so-called massive online open courses - MOOCs). Correlation analysis indicates that the three conversational agents could be safely used in specific automatic grading applications. More research is needed in the direction of system parameterization and also in the improvement of prompts (prompt engineering) to increase the accuracy and decrease the variability of the responses of these systems.