Μεταπτυχιακές Εργασίες

Μόνιμο URI για αυτήν τη συλλογήhttps://pyxida.aueb.gr/handle/123456789/51

Περιήγηση

Τώρα δείχνει 1 - 20 από 844

Artificial intelligence & health data in clinical trials leveraging innovation, exploring future opportunities & addressing emerging challenges
(2026-04-03) El-Malaouani, Antam; Ελ-Μαλαουάνι, Άνταμ; Chatziantoniou, Damianos; Spinellis, Diomidis; Μitrou, Lilian
Artificial Intelligence (AI) is increasingly integrated across all stages of the clinical trial lifecycle, supporting processes such as patient identification, feasibility assessment, risk monitoring, and protocol optimization. These developments contribute to improving the efficiency of clinical trials and enhancing evidence-based decision-making. However, the growing use of AI also introduces complex ethical, regulatory, and governance challenges. This thesis investigates the gap between existing ethical frameworks and their practical implementation in AI-enabled clinical trials. Through a structured review of interdisciplinary literature and policy documents, combined with the analysis of stakeholder-informed insights from pharmaceutical, regulatory, and digital health contexts, the study evaluates how AI-driven decision-support systems are governed in practice. The analysis focuses on transparency, accountability, data governance, human oversight, and explainability as key factors in maintaining institutional and public trust. The findings reveal a clear gap between formal ethical consensus and practical application. Although high-level principles are widely endorsed, their implementation remains uneven and often fragmented across organizational functions. Ethical governance is frequently approached as a compliance requirement rather than as an integrated component of trial design and execution. The study concludes that responsible AI adoption in clinical research requires a shift from principle-based ethics to embedded, lifecycle-oriented governance structures. By emphasizing the ethics-by-design approach, the thesis proposes practical pathways for bridging the gap between theoretical frameworks and real-world implementation.
The Load-Dependent Vehicle Routing Problem (LDVRP) (an optimization approach)
(2026-03-30) Dimoglou, Antonios; Δημόγλου, Αντώνιος; Chatziantoniou, Damianos; Zisis, Dimitrios; Zachariadis, Emmanouil
The Load-Dependent Vehicle Routing Problem (LDVRP) is a variant of the classical Capacitated VRP in which transportation cost depends not only on distance but also on the remaining load carried by the vehicle. This reflects real-world freight operations, where heavier vehicles consume more fuel and generate higher operational costs. This thesis develops a complete heuristic and metaheuristic framework for solving the LDVRP efficiently. The methodology begins with a Sweep heuristic that produces spatially coherent and capacity-feasible routes. This initial solution is refined using a load‑dependent Clarke–Wright Savings algorithm, followed by a multi-operator Local Search procedure employing 2‑opt, relocation, and swap moves to improve route structure and sequencing. To escape local optima and explore a wider solution space, an Adaptive Large Neighborhood Search with a Simulated Annealing acceptance criterion (ALNS‑SA) is implemented. The framework is evaluated on benchmark instances from the Golden dataset under different levels of load dependency. Results show that constructive heuristics significantly improve over geometric routing, while Local Search and ALNS provide additional gains through iterative refinement. The study demonstrates that incorporating load-dependent costs leads to more realistic routing decisions while remaining computationally tractable.
Ευρωπαϊκή δημόσια πολιτική: το νέο σύμφωνο μετανάστευσης και ασύλου
(2025-10-10) Αργυρίου, Ελευθερία-Κωνσταντίνα; Χριστόπουλος, Δημήτριος; Πεχλιβάνος, Λάμπρος; Μπλαβούκος, Σπυρίδων
Η παρούσα διπλωματική εργασία εξετάζει την εξέλιξη και τη δομή του Νέου Συμφώνου για τη Μετανάστευση και το Άσυλο, αναλύοντας τη μετάβαση από τις προγενέστερες πολιτικές στις μεταρρυθμίσεις που υιοθετήθηκαν το 2024. Μέσα από ποιοτική έρευνα πρωτογενών πηγών, διερευνάται η ιστορική πορεία της κοινής ευρωπαϊκής στρατηγικής και οι βασικοί πυλώνες του νέου πλαισίου, όπως η ενίσχυση των συνόρων και ο μηχανισμός αλληλεγγύης. Η μελέτη εστιάζει στα στάδια παραγωγής του Συμφώνου, από τις αρχικές προτάσεις έως τις προκλήσεις εφαρμογής τους στα κράτη μέλη. Τέλος, αξιολογείται η αποτελεσματικότητα του εγχειρήματος, επισημαίνοντας την πολυπλοκότητά του και την ανάγκη για ισχυρή πολιτική βούληση στην επίτευξη μιας ενιαίας ευρωπαϊκής πολιτικής.
Beyond isolated posts: persistent user dynamics in political misinformation on Reddit
(2026-03-10) Tsele, Anastasia; Τσελέ, Αναστασία; Kapetis, Chrysostomos; Chatziantoniou, Damianos; Kotidis, Yannis
This study investigates the diffusion of online misinformation by shifting the analytical focus from individual posts to user-level behavioral patterns. Using the FACTOID dataset, which contains millions of political posts from Reddit users, the analysis combines semantic text embeddings, psycholinguistic indicators derived from the LIWC framework, temporal activity patterns, and link-sharing behavior related to domain credibility and political bias. Semantic clustering techniques are applied to identify thematic communities, while supervised machine learning models including Logistic Regression, Support Vector Machines, and Random Forest, are used to classify users based on behavioral and linguistic characteristics. The findings demonstrate that behavioral signals, particularly the credibility of shared domains and repeated source-sharing patterns, are substantially stronger predictors of misinformation-related activity than emotional language indicators. In addition, the results reveal the presence of a persistent core of users who repeatedly activate during major political events, accompanied by waves of newly active accounts. Overall, the study suggests that misinformation should be conceptualized as a longitudinal behavioral dynamic rather than a series of isolated content incidents, highlighting the importance of user-level analysis for improving detection and governance strategies in digital information ecosystems.
Implementation of customer analytics: a comprehensive approach to sales force optimization, and recommendation-based sales growth
(2026-03-10) Mantaos, Dimitrios; Μαντάος, Δημήτριος; Chatziantoniou, Damianos; Zachariadis, Emmanouil; Poulymenakou, Angeliki
This essay briefly describes a business analytics project for BakConf SA, a company that operates in the bakery and confectionery industry, focusing on the transition from traditional decision-making and reporting to a data-driven approach. The project involves two ML models: one examines how sales field visits affect growth, and the other creates sales product suggestions based on the purchasing habits of top customers. Using the company’s data and analytics tools in R, this project provides practical insights for sales teams and managers. The reporting framework is designed to be clear and useful, helping the company quickly identify what works and where improvement is needed. Investing in analytics gives BakConf SA an advantage, making it easier to adapt to the market, connect with customers, and focus on growth. This aims to help BakConf SA increase its market share and raise the standard for how the industry operates, always aiming to be one step ahead of the competition.
Time series clustering: fuel price analysis across Greek counties
(2025-07-16) Xanalatou, Athina; Ξαναλάτου, Αθηνά; Ntzoufras, Ioannis; Chatziantoniou, Damianos; Karlis, Dimitrios
Time series clustering, as a valuable method in data mining and spatiotemporal analysis, has seen increasing relevance in economic studies, particularly in understanding regional market behaviors. This thesis applies such methodology to the case of retail fuel prices in Greece from 2022 to 2024, aiming to uncover latent structures across counties without relying on predefined labels. The adaptation of classical clustering algorithms to time-indexed price data necessitates a careful selection of similarity measures, where the challenge lies in balancing computational efficiency with sensitivity to the temporal dynamics of price evolution. Three clustering methods – K-Means, PAM with Euclidean distance, and PAM with Dynamic Time Warping – were evaluated across three fuel types. The results demonstrate that no single method universally outperformed the others. PAM with Euclidean distance produced the most cohesive and interpretable clusters for Diesel and Unleaded 98/100, capturing subtle spatial pricing regimes. In contrast, K-Means delivered slightly better internal cohesion and statistical separation for Unleaded 95, suggesting its effectiveness when price trajectories exhibit more synchronized temporal patterns. DTW-based clustering, while flexible in theory, underperformed in internal validation metrics and failed to distinguish groups clearly in most cases. The study identifies persistent regional disparities in fuel pricing, with notably higher costs in insular and peripheral areas, and highlights systematic price adjustments during national holidays and high-demand periods. These findings demonstrate the capacity of time series clustering to extract underlying economic patterns from complex and high-frequency price data. The results reinforce the value of unsupervised learning techniques in regional markets, while also emphasizing the methodological trade-offs between interpretability, flexibility, and computational scalability when applied to large-scale spatiotemporal datasets.
Beyond bitcoin: how different cryptocurrencies respond to country-specific geopolitical risks
(2026-02-02) Biliani, Ifigeneia; Μπιλιάνη, Ιφιγένεια; Lekakos, Georgios; Korfiatis, Nikolaos; Drakos, Konstantinos
This thesis explores the impact of geopolitical risk on cryptocurrency markets, focusing on eight digital assets, including major cryptocurrencies (Bitcoin, Ethereum, Ripple, Binance Coin, Tron) and stablecoins (Tether, USD Coin, Dai). It examines whether cryptocurrencies exhibit safe haven properties, how they respond to geopolitical risks from specific countries, and the dynamics of these responses over time. Key findings reveal that cryptocurrencies do not serve as safe havens during geopolitical uncertainty, as returns remain unaffected across all time horizons. However, geopolitical risk significantly influences trading volume, with effects peaking approximately four weeks after initial shocks. Stablecoins, particularly Dai and USD Coin, demonstrate heightened sensitivity to geopolitical risk, with Dai showing universal responsiveness across multiple countries. The study also highlights regional variations, finding that geopolitical risks originating from Russia, the U.S., and Israel have distinct impacts on specific cryptocurrencies. For example, Dai responds strongly to U.S. geopolitical risk, while Ethereum and Tether exhibit sensitivity to Israeli tensions. Bitcoin, in contrast, reacts to global geopolitical uncertainty but not to country-specific risks. Using advanced econometric methods, including lagged regressions and Vector Autoregression (VAR), the research identifies a one-month information diffusion window for geopolitical risk in cryptocurrency markets. The findings contribute to the understanding of cryptocurrency behavior during crises, offering insights for investors, policymakers, and developers on portfolio diversification, market stability, and risk management strategies. This thesis provides a comprehensive analysis of the complex and multidimensional relationship between geopolitical risk and cryptocurrency markets, addressing gaps in existing literature and paving the way for future research on digital asset behavior during global uncertainty.
Super app concepts: a strategic framework for platform-based innovation and digital service convergence in Greek retail banking: the Eurobank mobile app case
(2026-02-05) Miliotou, Eftychia; Μηλιώτου, Ευτυχία; Androutsos, Athanasios; Mitropoulos, Dimitrios; Kontopoulos, Ioannis
The digital banking landscape in Europe is currently undergoing a paradigm shift, transitioning from fragmented, single-utility applications toward integrated 'Super App' ecosystems. While the success of these models in Eastern markets was driven by rapid leapfrogging and limited regulation, the European trajectory is uniquely constrained by stringent data sovereignty (GDPR), competitive levelling (DMA), and a mature, multi-bank consumer base. To evaluate the viability of this transition, this research seeks to dissect the intersection of technical interoperability, consumer psychology, and economic sustainability. By examining how third-party integrations and 'open' service architectures influence the user journey, this study aims to define whether the banking Super App is a sustainable evolution of financial services or a structural overreach.
Agentic AI: business process automation in SMEs through artificial intelligence tools
(2026-02-05) Vasilopoulou, Eirini; Βασιλοπούλου, Ειρήνη; Sarantopoulos, Panagiotis; Poulymenakou, Angeliki; Doukidis, Georgios
This thesis examines the role of Artificial Intelligence (AI) in the automation of business processes in small and medium-sized enterprises (SMEs), with an emphasis on the use of Agentic AI and Retrieval-Augmented Generation (RAG) architectures. There is created a comprehensive Agentic AI system (technological artifact), inside an automation platform (n8n), employing a Design Science Research Methodology. The aim of the thesis is to transform the theoretical capabilities of modern ΑΙ systems into practical and functional digital tools capable of supporting core administrative functions of SMEs such as Sales, Marketing, Finance, and Administration, reducing the time spent on repetitive, multidimensional and cognitively demanding tasks. This method allows the thesis to address specific research questions such as designing and developing Agentic and Agentic RAG architectures, measuring the performance of AI agents through KPIs, and picturing AI agents as digital "role assistants" aiding different organizational functions. This thesis follows the Design Science Research (DSR) methodology, proposed by Peffers et al. (2007), with the aim of forming a structured methodological framework for the development of innovative Information Systems. This process allows for the systematic identification of the business problem, the definition of the design objectives, and the translation of the theoretical objectives into an applicable development framework, which is then used for the design, implementation, and evaluation of the proposed Agentic AI system. More specifically, the methodology is applied on a real operational context of a SME in the Used Car industry, which is used as a starting point for the analysis of the requirements for an AI system implementation. As a result of this process, a structured framework is developed that can be followed by a wide range of SMEs for the deployment of innovative Information Systems. Furthermore, following the Digital Transformation framework described by Doukidis et al. (2020), the research illustrates in what ways Agentic AI systems have the potential to facilitate Business Process Transformation and Organizational Transformation if intelligent agents are integrated into main managerial workflows. In addition, there are designed multiple specialized conversational AI agents based on RAG architectures, in order to connect with the company information, which operate autonomously to support Managerial decision-making across different Organizational Departments, such as Sales, Marketing, Finance, and Administration. Moreover, a central Management AI Agent is created, which serves as the only point of interaction for all the Managers, directing questions to the relevant functional agent and providing consistent and well-documented answers. Finally, there is carried out a quantitative performance evaluation of the system that demonstrates how Agentic RAG systems enhance not only operational efficiency but also decision-making quality.
Ανάλυση πιστωτικού κινδύνου με αλγορίθμους μηχανικής μάθησης και τεχνικές explainable AI. Έμφαση στη διαφάνεια, δικαιοσύνη και μεροληψία
(2026-02-18) Ζντράβα, Τζέσι; Ανδρουτσόπουλος, Κωνσταντίνος; Φούσκας, Κωνσταντίνος; Ζάρας, Ανδρέας; Λεκάκος, Γεώργιος
Ο σκοπός της παρούσας διπλωματικής εργασίας είναι η ανάπτυξη και εφαρμογή μιας μεθοδολογίας ανάλυσης πιστωτικού κινδύνου με τη χρήση αλγορίθμων μηχανικής μάθησης και τεχνικών Explainable AI, με έμφαση στη διαφάνεια, τη δικαιοσύνη και την μεροληψία στις αλγοριθμικές αποφάσεις. Η ανάλυση πιστωτικού κινδύνου είναι μια κρίσιμη διαδικασία των πιστωτικών ιδρυμάτων, καθώς επηρεάζει άμεσα τη λήψη αποφάσεων δανεισμού και συνδέεται με αυξημένη απαίτηση λογοδοσίας και συμμόρφωσης με το κανονιστικό πλαίσιο. Η μεθοδολογία εφαρμόστηκε στο German Credit Dataset και περιλαμβάνει τη συστηματική προεπεξεργασία των δεδομένων, την εκπαίδευση και σύγκριση πολλαπλών μοντέλων ταξινόμησης, καθώς και τη ρύθμιση υπερπαραμέτρων και την τελική αξιολόγηση σε ανεξάρτητο σύνολο ελέγχου. Η προβλεπτική απόδοση των μοντέλων αξιολογήθηκε με διάφορες μετρικές για την ανάλυση πιστωτικού κινδύνου. Επιπλέον πραγματοποιήθηκε και ανάλυση κατάταξης μέσω cumulative lift, ώστε να ελεγχθεί η ικανότητά τους να ιεραρχούνται οι δανειολήπτες με βάση τον εκτιμώμενο κίνδυνο. Εφαρμόστηκαν τεχνικές Explainable AI, όπως οι SHAP και LIME για να εξεταστεί γιατί το μοντέλο πήρε μια απόφαση και για να κατανοήσουμε τους παράγοντες που επηρεάζουν τις αποφάσεις του. Ζητήματα δικαιοσύνης και πιθανής μεροληψίας εξετάστηκαν με ερμηνευτικές τεχνικές και με ποσοτικές μετρικές fairness, δίνοντας έμφαση στον ρόλο ευαίσθητων δημογραφικών χαρακτηριστικών και στη σχέση τους με τις προβλέψεις των μοντέλων. Αυτή η έρευνα είναι σημαντική διότι προσφέρει μια συνοπτική αλλά ολοκληρωμένη εικόνα των τεχνικών και προκλήσεων που σχετίζονται με την ανάλυση πιστωτικού κινδύνου, ζήτημα που είναι ιδιαίτερα επίκαιρο στον χρηματοπιστωτικό τομέα. Στην εργασία συνδυάζονται η προβλεπτική απόδοση με τη διαφάνεια και την ερμηνευσιμότητα των μοντέλων, με σκοπό να καλυφθεί η αυξανόμενη ανάγκη για υπεύθυνη χρήση αλγοριθμικών συστημάτων από τα πιστωτικά ιδρύματα. Ακόμη, οι πρόσφατες κανονιστικές εξελίξεις, όπως η εισαγωγή του AI Act και του GDPR, έχουν αναδείξει την κρίσιμη ανάγκη επεξηγησιμότητας και λογοδοσίας στην εφαρμογή τεχνικών μηχανικής μάθησης στην πιστοληπτική αξιολόγηση.
Evaluation of static analysis tools for detecting XSS vulnerabilities in PHP code: a comparative study using the CrossVul dataset
(2026-02-01) Tsokanaridis, Athanasios; Τσοκαναρίδης, Αθανάσιος; Koniakou, Vasiliki; Kontopoulos, Ioannis; Mitropoulos, Dimitrios
This thesis presents a comparative evaluation of three distinct static analysis paradigms for detecting Cross-Site Scripting (XSS) vulnerabilities in PHP applications: taint tracking (Psalm), heuristic-based pattern matching (Semgrep), and type-based analysis (PHPStan). Using the Cross Vul dataset of real-world vulnerable code fragments, the study establishes a reproducible methodology to measure key performance metrics, including precision, recall, and F1 score. The results indicate that no single paradigm consistently dominates across all evaluation metrics, reflecting a structural trade-off space in vulnerability detection. PHPStan (augmented with a custom rule) achieved the highest recall (0.2396) and F1 score (0.3244) but produced the highest false positive rate. Psalm proved to be the most conservative and reliable tool, with the highest precision (0.6283). Semgrep occupied an intermediate position, excelling at identifying explicit insecure coding idioms but limited by its lack of global data-flow awareness. Qualitative analysis further identifies recurring failure patterns, such as difficulties in tracking data across persistence boundaries (stored XSS) and context-blindness in HTML attributes. The study concludes that effective XSS detection is best achieved through complementary tool integration rather than reliance on a single analysis strategy.
A structured methodology for AI business process automation in SMEs
(0031-01-26) Bounti, Antigoni-Iosifina; Μπούντη, Αντιγόνη-Ιωσηφίνα; Sarantopoulos, Panagiotis; Korfiatis, Nikolaos; Doukidis, Georgios
Digital Transformation is a strategic priority for modern businesses, as it affects not only the adoption of new technologies but also the way in which organizations redesign business processes, structures, and decision-making practices. In this context, Artificial Intelligence is recognized as a key catalyst for digital transformation. However, in small and medium-sized enterprises (SMEs), its use, and in particular its application in business process automation, remains limited, fragmented, and often unsustainable. The main obstacles are related to limited resources, low data maturity, lack of specialized skills, organizational uncertainty, and the absence of structured methodological guidance. This thesis aims to fill this gap by proposing and evaluating a structured methodology that supports digital transformation using artificial intelligence in small and medium-sized enterprises, with an emphasis on business process automation. Following the Design Science Research approach, the thesis develops the AIM-SME framework, a five-phase methodological artefact that guides companies from assessing organisational readiness and identifying appropriate processes to designing, piloting, integrating and continuously improving artificial intelligence solutions. The proposed framework is process-centric, gradual, and technologically neutral, treating the limitations of SMEs as basic design assumptions rather than secondary obstacles. The application of the methodology is presented through an empirical case study in a Greek company, highlighting how the structured approach supports evidence-based decision-making, prevents premature or inappropriate adoption of artificial intelligence, and enhances learning and strategic alignment in the context of digital transformation. Finally, the evaluation of the subsequent phases of the methodology is carried out through feedback from experienced executives and consultants in digital transformation and artificial intelligence, confirming the realism, usability, and transferability of the framework. The work contributes both theoretically, by enriching the literature on digital transformation and the adoption of artificial intelligence in small and medium-sized enterprises, as well as practically, by providing an applicable tool to support sustainable and gradual organizational transformation.
Mobile banking app development
(2026-02-05) Vervainiotis, Antonios-Panagiotis; Βερβαινιώτης, Αντώνιος-Παναγιώτης; Mitropoulos, Dimitrios; Sarantopoulos, Panagiotis; Kontopoulos , Ioannis
This thesis presents the design, development, and evaluation of a cross-platform banking branch management application aimed at improving the efficiency, accuracy, and usability of internal banking operations. While modern banks heavily invest in customer-facing digital services, internal administrative systems often remain fragmented, outdated, and difficult to maintain. This study addresses that gap by proposing a unified mobile and web-based solution for managing branch-level data. The proposed system is implemented using Ionic React for the frontend, FastAPI as a backend application layer, and Supabase as a Backend-as-a-Service (BaaS) platform providing database and authentication services. OpenStreetMap and the Nominatim API are integrated to support address search and geolocation, reducing manual data entry and improving data accuracy. The application is deployed both as a web application and as an Android mobile application through Capacitor, enabling code reuse across platforms. A design science research methodology is adopted, encompassing requirements analysis, system design, implementation, and evaluation. Usability testing was conducted with 18 participants representing diverse technical backgrounds. The evaluation employed task-based analysis, qualitative feedback, and the System Usability Scale (SUS). The results indicate high task completion rates and an average SUS score of 82.4, reflecting excellent perceived usability and user satisfaction. The findings demonstrate that modern hybrid application frameworks, combined with cloud based backend services and open geolocation platforms, can effectively support internal banking processes. The thesis contributes a reference architecture and empirical usability evidence for future research and practical implementations in administrative and enterprise focused financial applications.
Η είσοδος της τεχνολογίας στη δημόσια διοίκηση: η περίπτωση της χρήσης της τεχνητής νοημοσύνης (AI) στο Ελληνικό Κτηματολόγιο
(2025-12-20) Κουντουριώτη, Δήμητρα; Δενδραμής, Ιωάννης; Μπλαβούκος, Σπυρίδων
Η παρούσα έρευνα εστιάζει στην εισαγωγή της τεχνολογίας στο Ελληνικό Δημόσιο και πιο συγκεκριμένα στην παροχή των υπηρεσιών του Ελληνικού Κτηματολογίου. Ουσιαστικά, δίνεται έμφαση στην τεχνολογική έκρηξη που έχει σημειωθεί κατά την τελευταία δεκαετία και έχει οδηγήσει με τη σειρά της σε αρκετές αλλαγές στον τρόπο με τον οποίο λειτουργεί και διοικείται ο δημόσιος τομέας στην Ελλάδα. Θίγονται ηθικά ζητήματα που ανακύπτουν από την εκτεταμένη χρήση της τεχνητής νοημοσύνης για την παροχή υπηρεσιών προς τους πολίτες όπως για παράδειγμα ο ρόλος του ανθρώπινου δυναμικού, η προστασία των προσωπικών δεδομένων των χρηστών καθώς και η ποιότητα των παρεχόμενων υπηρεσιών. Ουσιαστικά, μέσα από την εκτεταμένη βιβλιογραφική ανασκόπηση, αλλά και τη χρήση της ποιοτικής έρευνας εξάγονται σημαντικά συμπεράσματα για το τι ακριβώς σημαίνει η τεχνητή νοημοσύνη για τους χρήστες και για τους πολίτες αλλά και για τον τρόπο με τον οποίο αυτή η νέα πραγματικότητα δύναται να αλλάξει την καθημερινότητα των πολιτών.
Resource allocation models in project management
(2026-02-05) Triantopoulou, Vasiliki; Τριαντοπούλου, Βασιλική; Lekakos, Georgios; Diakonikolaou, Kyriakos; Androutsopoulos, Konstantinos
Effective resource allocation is a critical success factor in project management, particularly in environments where resources are limited and projects are characterized by increasing complexity and uncertainty. Although a wide range of resource allocation methods has been developed in the academic literature, including resource leveling, heuristic approaches, and optimization models, their practical adoption in real organizational settings remains limited. This thesis focuses on human resource allocation, specifically the assignment of project managers across multiple projects within an organization, while considering multiple criteria and operational constraints. The objective is to adapt and apply a resource allocation model that supports balanced workload distribution, aligns managerial skills and experience with project requirements, and enhances overall organizational effectiveness. The study combines a focused literature review with a case study inspired by a real organizational environment in the banking sector. Using an optimization-based approach implemented through Excel Solver, different allocation strategies are evaluated in terms of assignment suitability, workload balance, and time feasibility. The findings indicate that multi-criteria resource allocation models can significantly support more informed project staffing decisions, helping bridge the gap between theoretical approaches and the practical needs of modern organizations.
Leveraging digital technologies to create behavioural incentives: an investigation into the technological and business challenges associated with the use of digital technologies to encourage and drive the adoption of specific user behaviours in automotive e-commerce
(2026-02-03) Komselis, Theofanis; Κομσελής, Θεοφάνης; Lekakos, Georgios; Nikolaou, Ioannis; Fouskas, Konstantinos
This thesis investigates the impact of digital nudges on user behaviour in automotive e-commerce. Drawing on Nudge Theory, the Fogg Behavior Model, and Cialdini's principles of persuasion, the research develops a taxonomy of eight digital nudge categories and examines their relationship with behavioural metrics. The methodology comprises a mixed-methods case study incorporating UX audit and Google Analytics 4 data analysis (11.5 million page views, 14,266 conversions). Statistical analysis was conducted using ANOVA and t-tests. Findings reveal statistically significant differences in conversion rates across page types (F=193.158, p<0.001). Pages with concentrated trust nudges achieve 8.9 times higher conversion rates, while traffic source affects effectiveness by a factor of 52. The research contributes theoretically by introducing the concepts of stage-specific nudge effectiveness and nudge concentration, extending the Fogg Model with an intent-nudge alignment framework. Practically, findings indicate the need to tailor nudges to customer journey stages and segment based on user intent.
CharonXray: hybrid detection of vulnerabilities in Python native extensions
(2026-02-05) Papadongona, Maria; Kontopoulos , Ioannis ; Kechagia, Maria; Mitropoulos, Dimitrios
Τα σύγχρονα συστήματα Python βασίζονται ολοένα και περισσότερο σε εγγενείς επεκτάσεις γραμμένες σε C/C++ για να πετύχουν αυξημένη απόδοση και λειτουργικότητα πέρα από την Python. Ωστόσο, οι εγγυήσεις ασφάλειας της Python δεν ισχύουν για τις εγγενείς επεκτάσεις, επανεισάγοντας κινδύνους ασφάλειας μνήμης και άλλες ευπάθειες χαμηλού επιπέδου. Έτσι, ευπάθειες μπορεί να προέρχονται από εγγενή κώδικα αλλά να είναι προσβάσιμες μέσω φαινομενικά ασφαλών Python διεπαφών. Η ανίχνευσή τους απαιτεί ανάλυση που υπερβαίνει τα όρια γλωσσών και μοντελοποιεί τις γέφυρες μεταξύ καλέσιμων οντοτήτων της Python και υλοποιήσεων σε εγγενή κώδικα. Η παρούσα διπλωματική εργασία εξετάζει ευπάθειες που προκύπτουν από αλληλεπιδράσεις Python και εγγενή κώδικα, αναδεικνύοντας περιορισμούς της στατικής ανάλυσης, η οποία συχνά αποτυγχάνει να εντοπίσει δυναμικά δημιουργούμενες γέφυρες κατά τον χρόνο εκτέλεσης. Αυτό οδηγεί σε εσφαλμένη ταξινόμηση ευάλωτων εγγενών διαδρομών ως μη προσβάσιμων και σε ελλιπή κάλυψη. Για την αντιμετώπιση του προβλήματος, συνδυάζεται το στατικό εργαλείο διαγλωσσικής ανάλυσης CHARON με το δυναμικό εργαλείο PyXray, το οποίο ανακτά τις γέφυρες μέσω ανάλυσης της δομής των αντικειμένων κατά τον χρόνο εκτέλεσης. Το CharonXray συνδυάζει τη στατική διαγλωσσική ανάλυση με δυναμικά ανακτημένες γέφυρες, βελτιώνοντας τη μοντελοποίηση εγγενούς προσβασιμότητας. Η προσέγγιση αξιολογήθηκε σε μεγάλο σύνολο πραγματικών Python πακέτων με εγγενείς επεκτάσεις. Τα αποτελέσματα δείχνουν ότι το PyXray εντοπίζει σημαντικά περισσότερες γέφυρες (57,5% περισσότερα πακέτα) και ότι η ενσωμάτωσή τους αυξάνει τα πακέτα και τις ροές εγγενών ευπαθειών που εντοπίζονται. Το CharonXray αποκάλυψε νέες ευπάθειες σε ευρέως χρησιμοποιούμενα πακέτα, όπως aubio, autobahn-python, aioquic και pymunk, συμπεριλαμβανομένων διαρροών μνήμης, υπερχειλίσεων ακεραίων, σφάλματα διαίρεσης με το μηδέν και αποαναφορές NULL δεικτών. Συμπερασματικά, η υβριδική ανάλυση είναι κρίσιμη για ακριβή μοντελοποίηση διαγλωσσικών επιφανειών επίθεσης σε σύγχρονο διαγλωσσικό λογισμικό.
Development and implementation of a methodological framework for evaluating critical success factors in large supply chain projects
(2026-02-03) Galanaki, Angeliki; Γαλανάκη, Αγγελική; Lekakos, Georgios; Manolopoulos, Dimitrios; Androutsopoulos, Konstantinos
This thesis focuses on the identification, evaluation, and prioritization of Critical Success Factors (CSFs) that influence the successful implementation of large-scale supply chain projects in Fast Moving Consumer Goods (FMCG) companies. In today’s dynamic and highly competitive business environment, FMCG organizations are required to manage multiple and diverse supply chain projects simultaneously, ranging from infrastructure and operational initiatives to digitally enabled transformation projects. Within this context, understanding which factors most strongly drive project success is essential for effective decision-making and resource allocation. To address this objective, the study develops and applies a methodological framework based on the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS). Thirteen Critical Success Factors were identified through an extensive literature review and subsequently evaluated using a five-point Likert scale by experienced professionals involved in supply chain projects at a multinational company. The empirical analysis examined four distinct supply chain projects. The quantitative findings are further enriched by qualitative insights obtained through semi-structured interviews, which emphasize the role of practical management mechanisms such as maintaining a strong project “drumbeat,” effective communication routines, and continuous monitoring of benefits. Together, these results suggest that a one-size-fits-all approach to supply chain project management is ineffective. Instead, FMCG companies should tailor governance structures, managerial focus, and resource allocation according to the specific characteristics of each project. Overall, this thesis contributes to both academic research and managerial practice by proposing a structured, data-driven approach for evaluating critical success factors in supply chain projects.
The impact of AI-generated advertisements on customers’ perceptions
(2026-02-05) Veneva, Desislava Petrova; Kechagia, Maria; Doukidis, George I.; Sarantopoulos, Panagiotis
The thesis investigates how the disclosed creator of an advertisement, human versus artificial intelligence (AI), influences consumers’ perceptions of advertising creativity. Drawing on Attributional Theory of Creativity, the study examines whether consumers attribute different levels of creativity to advertisements based solely on the perceived source of creation, even when the advertising stimulus remains identical. In particular, the research focuses on the mediating role of perceived emotionality of the creator in shaping creativity evaluations. To address the research objectives, a quantitative, experimental research design was employed using a between-subjects online survey. Participants were randomly assigned to view a static digital advertisement labeled as either human-created or AI-generated, accompanied by a manipulation text reinforcing the creator identity. Perceived advertising creativity and perceived emotionality of the creator were measured using validated scales. The results indicate that advertisements labeled as human-created are perceived as significantly more creative than identical advertisements labeled as AI-generated. Furthermore, perceived emotionality of the creator fully mediates the relationship between advertisement creator and perceived advertising creativity, suggesting that consumers’ creativity judgments are driven by emotional attributions rather than the creator label alone. The study contributes to the growing literature on AI creativity and advertising by empirically demonstrating the persistence of human superiority biases in creative evaluations and highlighting emotionality as a key explanatory mechanism. From a managerial perspective, the findings suggest that transparency about AI usage in advertising should be handled strategically, as AI disclosure may influence consumer perceptions beyond objective creative quality. On the whole, this thesis advances theoretical understanding of creativity attribution in the age of AI while offering practical insights for advertisers navigating human–AI creative boundaries.
Authentic vs inauthentic artificial empathy in customer-service chatbots: effects on user’s satisfaction
(2026-02-03) Filippaiou, Elpida; Φιλιππαίου, Ελπίδα; Nikolaou, Ioannis; Lekakos, Georgios
This research investigates when users perceive artificial empathy in chatbots as authentic versus inauthentic, and how these perceptions affect their experience and satisfaction in customer service. To discover this question, a controlled online experiment was conducted that tested three different chatbot scenarios, always using the same conversation framework with the user. There were three types of chatbots: the mechanical one, which was task-focused; the empathetic one, which used emotional language appropriate to the situation; and the inauthentic empathetic one, which used exaggerated, overly enthusiastic emotional language. After randomly participating in one of the scenarios, participants completed a questionnaire that used a seven-point Likert scale to measure perceived authenticity, perceived social presence, and user satisfaction. The analysis incorporated group-level comparisons with a mechanism-oriented path framework, estimated via regression-based mediation and bootstrapped indirect effects. In all conditions, perceived social presence was the dimension that was consistently distinguished, while perceived authenticity and satisfaction were less distinguishable at the average level. The experiment ultimately showed that the type of chatbot was not directly linked to satisfaction and that perceived authenticity did not affect the relationship between the type of chatbot and satisfaction, as was the initial hypothesis. Instead, the results showed that a more indirect mechanism seemed to be at work. Perceived authenticity was closely linked to perceived social presence, which then predicted satisfaction. The results showed that simply demonstrating empathy was not enough to improve the user experience. In the end, it seemed that the best way to achieve higher satisfaction rates was to design interactions that maintained social presence by using empathetic language that was relevant to the user's situation while also keeping the work moving forward.

Περιήγηση

Πρόσφατες Υποβολές