CONTENTS
|
RPI: An Object-Oriented Interpreter for Romanian-Language
Pseudocode in Computer Science Education Valentin-Ștefan ȘANDOR, Bogdan IANCU 5 The current paper presents the steps used and the development process of a custom interpreter for the Romanian pseudocode language that offers native Object-Oriented Programming (OOP) support. A custom scanner was designed to be able to tokenize the Romanian keywords, and the results were fed forward to a parser that constructs an Abstract Syntax Tree (AST). The AST is then evaluated by an interpreter that executes the instructions. Advanced features such as recursion, method overriding, and user-friendly error messages are also supported. The trials showed great potential for in-class usage and can make the interpreter a stepping-stone for introducing newer paradigms to students, as is the case of OOP. Future work in-cludes debugging support, which can be easily developed since the codebase was designed with a focus on extensibility. Keywords: Pseudocode, Interpreter, Romanian, Computer science, Education Artificial Intelligence Agents in ERP Ecosystems: An Analysis of the Gap Between Technological Discourse and Actual Integration Maturity Octavian DOSPINESCU, Cristian SIMIONESCU, Daniel-Florin DĂNILOAIA 20 Enterprise Resource Planning (ERP) systems serve as the operational backbone of modern companies, representing a field undergoing continuous technical and operational transformation and improvement. Although technological realities necessitate the modernization and integration of new technologies into ERP systems, the integration of autonomous artificial intelligence (AI) agents into these ecosystems currently lags significantly behind the marketing rhetoric surrounding the field. This article aims to highlight and analyze the discrepancy between the "agentic" rhetoric promoted by ERP vendors and the actual maturity level of autonomous functionalities implemented in market-leading platforms. Based on a systematic review of the literature from the past decade, we formulated three research hypotheses regarding: (H1) the difference between the prevalence of conversational AI assistants and that of AI agents with genuine transactional autonomy; (H2) the relationship between an ERP platform's cloud-native architecture and the maturity level of its integrated AI agents; and (H3) the gap between the intensity of "agentic" marketing rhetoric and the documented governance levels of these systems. The hypotheses were tested using a quantitative content analysis applied to the public documentation of 12 internationally representative ERP platforms. The analysis employed a coding instrument structured around four dimensions: the presence of conversational assistants within the ERP ecosystem, actual agentic maturity, the intensity of marketing rhetoric, and documented governance. The results of our objective analysis confirm the existence of a statistically significant gap (p=0.002) between truly autonomous functionalities and those that are merely assistive. At the same time, the analysis rejects (p=0.433) the existence of a significant association between cloud-native architecture and agentic maturity. Furthermore, our findings (p<0.001, mean gap = 1.44 points on a normalized 5-point scale) indicate a significant discrepancy between documented governance and marketing discourse. Our article makes a significant contribution to the study of agentic AI within the context of ERP systems, while also offering recommendations for managerial practice and potential avenues for future research. Keywords: Agentic AI, ERP systems, Artificial intelligence agents, Digital transformation, Autonomous automation A Hybrid Review of Supervised Machine Learning for Banking Transaction Fraud Detection: A Bibliometric and PRISMA-Based Analysis Nguyen Thi HANG 33 This study provides a systematic review of research on the application of supervised machine learning to banking transaction fraud detection, aiming to identify gaps in the existing literature. The study integrates bibliometric analysis using the bibliometrix package in R and VOSviewer with the PRISMA screening method. The dataset comprises 292 Scopus-indexed publications for bibliometric analysis and 171 articles meeting PRISMA criteria for qualitative synthesis. Results reveal significant growth in the field, dominated by decision tree–based models and ensemble techniques. However, four key gaps are identified: (i) research bias toward credit card fraud; (ii) over-reliance on binary classification; (iii) predominant use of traditional SMOTE to address data imbalance; and (iv) limited feature interpretability, particularly regarding the direction of feature effects and behavioral attributes on predicted risk. The study recommends expanding research to debit and mobile banking fraud, developing risk-score models, and evaluating SMOTE variants with concept drift handling to enhance transparency. Keywords: Banking Transaction Fraud, Bibliometric Analysis, PRISMA, Supervised Machine Learning The Expert Systems as an Artificial Intelligence Solution for E-Commerce: Outlining a Conceptual Framework Natalia BURLACU 51 The increasing integration of Artificial Intelligence (AI) and digital technologies in e-commerce requires intelligent solutions capable of supporting decision-making, improving customer experience, and optimizing business processes. However, existing research mainly addresses specific AI applications, while limited attention has been paid to comprehensive frameworks for the design of Expert Systems (ES) adapted to e-commerce environments. This research proposes a conceptual and methodological framework for the design, development, and implementation of ES-based solutions in e-commerce. Based on a critical review of the specialized literature and comparative analyses of existing approaches, including machine learning, deep learning, and large language models, the study identifies the current limits of existing research in the field and defines guidelines for the further development of ES adapted to the requirements as well as the possibilities of new generations of technologies. The proposed framework integrates com-putational, economic, and human-centered perspectives, contributing to a structured methodology for future intelligent, expert system-type applications orientated towards the e-commerce market. This paper is intended for researchers, software developers, and practitioners interest-ed in e-commerce digitalization and intelligent solution development. Keywords: Application of Expert Systems (ES), AI E-Commerce Solution, Conceptual Study, AI Subdomains Representation, ES vs AI approaches Rethinking EdTech: A Multi-Criteria Comparative Analysis of EdTech Platforms Oana-Larisa STOICA, Andrei-Cătălin NICA 67 The rapid evolution of artificial intelligence (AI) is transforming Educational Technology (EdTech), enabling new forms of personalized learning, real-time feedback and adaptive instruction. Despite this potential, the ways in which AI is integrated into learning platforms vary significantly, affecting their effectiveness, usability and transparency. This research investigates these differences through a multi-criteria evaluation of AI-driven EdTech platforms, with a comparative focus on Romanian and international solutions. Six categories of platforms are examined, using a mixed-methods approach that combines qualitative functional analysis, semi-quantitative scoring and comparative assessment. A multi-criteria decision analysis (MCDA) model is applied, evaluating five dimensions: AI functionalities, personalization, feedback quality, user experience and transparency. The results show that platforms with advanced AI integration, such as adaptive learning and intelligent tutoring systems, achieve higher levels of personalization and feedback quality but often at the expense of transparency. This study offers a novel evaluation framework, a functional taxonomy of AI in EdTech and an evidence-based comparison, providing valuable insights for educators, developers and policymakers aiming to advance AI-enhanced learning environments. Keywords: Educational Technology Platform, Artificial Intelligence, Learning, Adaptivity, Personalization Detecting AI-Generated Product Reviews Using Classical Machine Learning Techniques Sebastian TONU, Ioana-Alexandra TONU 81 The increased use of online product reviews has made them a key factor in consumer decision-making processes within e-commerce platforms. In this context, recent advances in large language models (LLMs) have enabled the generation of artificial reviews that closely resemble human-written content, raising concerns regarding their impact. This paper proposes a practical approach for detecting AI-generated product reviews based on classical machine learning techniques. A dataset is constructed by combining real user reviews with content generated us-ing a large language model, followed by a preprocessing stage for data standardization. The classification task is addressed using TF-IDF features and a linear Support Vector Machine model. In addition, an explainability component based on SHAP is employed to analyze the contribution of individual linguistic features to the model predictions. The experimental results indicate that the proposed approach achieves high classification performance while also providing interpretable insights into the patterns that differentiate generated and authentic reviews. Keywords: Fake reviews, LLM, TF-IDF, Support Vector Machine, Explainable AI Publishing Guide for Authors 94 INFOREC Association 96 |
