This paper presents a metadata analysis framework to investigate research trends and predict citations in the field of recommender systems. We apply Natural Language Processing (NLP) techniques, using spaCy and regex, to extract and analyze frequencies of different system types and technologies. To explore relevant researchtopics, we implement both BERTopic and Non-negative Matrix Factorization (NMF). We also introduce a zeroshot classification pipeline with the facebook/bart-large-mnli model to categorize articles into recommender system categories without requiring supervision. For citation prediction, we employ the distilbert-base-uncased model from Hugging Face to tokenize and vectorize text features, engineering 13 additional features to enrich the dataset. Random Forest, XGBoost and LightGBM models are used to evaluate predictive performance using RMSE and R² under both baseline and feature-enhanced scenarios. The proposed feature engineering and dimensionality reduction approach significantly improve citation prediction performance compared to the baseline. Specifically, RMSE is reduced from approximately 19-21 to around 6.8-10.6, while the coefficient of determination (R²) increases from 0.38-0.45 to up to 0.93 across the evaluated models. The most notable improvement is observed for the Random Forest and LightGBM models, where RMSE decreases by over 65% and R². Context-aware systems, with approximately 300 mentions and knowledge-based systems, with fewer than 100 mentions, receive comparatively less attention in the literature. BERTopic reveals a diverse and evolving landscape in recommender system research, spanning deep learning models, privacy and federated learning, domain-specific applications in tourism, education, healthcare and fake news. Explainable and cross-domain recommendations appear more prominently in recent years.

The purpose of our paper is to develop a unified multi-agent architecture that automates end-to-end machine learning (ML) pipeline generation from datasets and natural-language (NL) goals, improving efficiency, robustness and explainability. A five-agent system is proposed to handle profiling, intent parsing, microservice recommendation, Directed Acyclic Graph (DAG) construction and execution. It integrates code-grounded Retrieval-Augmented Generation (RAG) for microservice understanding, a Pareto-based multi-objective recommender that combines keyword relevance, semantic similarity, data compatibility, execution-history evidence, an adaptive learning from execution history and a self-healing mechanism using Large Language Model (LLM)-based error interpretation. The approach is evaluated on 150 ML tasks across diverse scenarios. The system achieves an 84.7% end-to-end pipeline success rate, outperforming baseline methods. It demonstrates improved robustness through self-healing and reduces workflow development time compared to manual construction. The paper introduces a novel integration of code-grounded RAG, explainable recommendation, self-healing execution and adaptive learning within a single architecture, showing that tightly coupled intelligent components outperform isolated solutions.

Writing content with the assistance of artificial intelligence has increased in popularity and influenced every domain, yet the challenge of understanding the differences between machine-generated and human-authored writing remains. To address this issue, our research aims to propose a detailed framework for deciphering the characteristics of machine-written language by identifying distinct linguistic features, analyzing stylistic differences, evaluating sentiment consistency, and validating the performance of selected generative models in replicating human writing across different types of writing. Using free resources from Hugging Face, we selected three models: GPT-Neo 1.3B, Qwen2.5-1.5B-Instruct, and BloomZ-560M, and employed them without fine-tuning in a Google Colab environment using a consistent prompt. These models were used to generate introductions for corresponding human-authored texts. We compared the extracted features of 2,121 texts, using visualisation methods such as histograms, boxplots and Q-Q plots, alongside the Shapiro–Wilk test and the non-parametric Kruskal – Wallis test and Dunn’s post-hoc test. Our findings highlight both the advancements of open-source text-generating models and the persistent gaps in replicating the depth and richness of human writing, providing and contributing to the knowledge about artificial intelligence and its progress in the generative field, supporting the validation of open-source text-generative models through statistical comparisons.

Information is ubiquitously available at our fingertips, transforming the way we learn, work and engage with the world around us. The challenge is not just accessing data but discerning its relevance and utility. This constant flow of information demands selective attention and strategic thinking about how we integrate new findings for professional growth. In this context, we propose an e-learning platform that recommends career paths based on user-uploaded PDFs. Our solution extracts keywords with Natural Language Processing (NLP). Using OpenAI, we enable interaction with the PDF files, allowing the user to ask questions and receive summaries. Then, we generate embeddings and index them with Facebook AI Similarity Search (FAISS). Next, we use a dataset of job listings and, with BERT, skills and technologies are extracted. An interconnected graph using a graph database system (Neo4j) based on these skills and technologies is built. Keywords from the uploaded documents are analyzed and matched to skills, leading to job recommendations or guidance on additional skills needed to secure employment. Mean Reciprocal Rank (MRR) is calculated to compare the results of different job recommendation systems.

This paper examines the delicate balance between fake news and freedom of speech by analyzing academic research at the intersection of “fake news” and “election.” Using 568 publications from the Web of Science database, the research applies data analytics and natural language processing (NLP) methods to uncover geographical patterns, thematic trends and sentiment in the scholarly discourse. Named Entity Recognition identifies the most studied countries and platforms, while topic modeling reveals recurring themes such as political misinformation, media influence and the role of artificial intelligence in fake news detection. The originality of this work lies in its integration of bibliometric analysis with advanced NLP-driven semantic, large language models and sentiment approaches, offering a multidimensional view of how fake news is framed in academic contexts. Unlike prior studies, it quantifies emerging trends using Compound Annual Growth Rate (CAGR), identifies propagation mechanisms like “sharing” and “engagement,” and bridges computational and social science perspectives. Findings show a strong research focus on misinformation during elections since 2016, with Brazil, Spain and the United States as leading contributors, and Twitter® and Facebook® as the most frequently studied platforms. Our research contributes to a deeper understanding of how misinformation and freedom of speech are debated in scholarly literature and underscores the need for interdisciplinary approaches to safeguard democratic integrity in the digital era.

Predicting the willingness and motivating the prosumers to participate in P2P markets is a challenging issue; thus, this paper proposes a methodological framework to address it. Valuable insights are extracted from user preferences, such as the most preferred technology and the motivation to participate in P2P markets. To assess the benefits of participating in LEM, a trading environment is simulated using 114 apartments and a photovoltaic (PV) system of 300 kWp, and several Key Performance Indicators (KPI) are calculated. Then, to increase the self-sufficiency of the market, the selling prices on LEM are used to motivate participants to shift their load during the trading hours. A load optimization algorithm is proposed to minimize the daily cost considering the selling prices and trading probability on LEM. In case all participants are optimizing their load, more savings and greater environmental benefits are obtained. The simulation of LEM trading among 114 households using a shared 300 kWp PV system demonstrates significant improvements in both economic and energy self-sufficiency KPIs. Without optimization (Case 1), annual savings reached €9,709 with a 5% increase in SSI and 12% in SCI. When half of the participants shifted their load (Case 2), savings increased to €12,340 and SCI rose to 92%. With full optimization (Case 3), annual savings peaked at €12,921, SCI reached 99%, and the average cost per member dropped by €266.3. A further 10% increase in PV capacity led to €41,205 in savings and improved GDI by 3%. These results confirm that LEM incentives and demand-side flexibility significantly enhance local RES utilization, reduce grid reliance and enable reinvestment in sustainable infrastructure.

Home assistants gained more attention as they provide guidance and can also be active in energy communities and assist prosumers by controlling appliances. Graphical interface and interaction with appliances via vocal commands have transformed home assistants into valuable tools. In this paper, we propose a methodology embedded within an Energy Assistant (EA) that incorporates algorithms for forecasting generation availability, recommending optimal time slots for load scheduling, optimally scheduling appliances, setting trading strategies in Local Electricity Markets (LEM), and controlling the operation of appliances. EA provides decision support for trading on LEM using reinforcement learning agents trained to optimize the price and quantities to maximize the trading probability and minimize the cost. The EA algorithms are embedded as a home assistant into an Alexa skill that has multiple intents corresponding to user requests. For simulations, 114 apartments, 7 houses with small Photovoltaic (PV) systems and a large PV system of 500 kW are considered. When 51 % of the households use EA, Self-Consumption Index increases from 0.72 to 0.85 and Grid Dependence Index reduces from 56 to 32. The average annual cost per apartment decreased from €96.62 to €76.60 after using EA for trading, indicating savings of 37.80 %.

Abstract Cryptocurrencies are in the center of attention of investors, public authorities and researchers, but the interest has shifted from purely financial aspects regarding the way of trading, lack of regulation and supervision of transactions, volatility, correlation with other assets to aspects related to sustainability taking in account the high energy consumption generated by the mining process and the impact on environmental pollution. Bitcoin was chosen for the research considering the dominance that this financial asset has on the cryptocurrency market and its position as alpha currency.The article focuses on the relationship between Bitcoin transactions and energy consumption, for period 1st January 2019—31st of May 2022, this interval having significant price movements. The authors made a prediction of the Bitcoin price using a complex meta-model and SQL analytical functions. The analysis is based on 15 fundamental variables in order to forecast the price: Bitcoin data (prices and volume), electricity price and traded quantity on day-ahead market (DAM), gas price and traded quantity on DAM, inflation in EU, EU-ETS emissions certificates and oil prices. The study reveals the importance of the relationship Bitcoin—energy—carbon emissions, elements that capture the impact of the mining process on the environment from the perspective of energy consumption. Investors on the Bitcoin market must be aware not only of the importance of financial aspects on the price of cryptocurrencies (inflation, demand, offer), but also of other elements related to the evolution of energy prices (electricity, oil, gas, renewable energy) and the evolution of emissions certificates prices. Considering the promotion of the principles of sustainable development on the capital market, portfolio investors have become increasingly attentive to the social and environmental performance of financial assets. This study aims to make financial market players aware of the non-financial implications of their transactions. In addition, the energy transition and the reconfiguration of the energy mix are elements of impact on the cryptocurrency market through the technical levers involved in the mining process.

We investigate the performance of intelligent systems such as various Long Short-Term Memory (LSTM) and hybrid models to forecast the electricity spot prices considering univariate and multivariate models. Six models are created to handle the Electricity Price Forecast (EPF). Furthermore, an EPF methodology that consists of a LSTM univariate model, namely Single in–out (Sio) model is proposed. It builds on the Day-Ahead electricity Market (DAM) specificity and, as a novelty, it inserts the predicted value back into the sliding input vector to predict the next values until the entire vector of 24 prices is predicted. The proposed model is further enhanced by the convolutional reading of input data that is embedded into the LSTM cell or by a hybrid combination of LSTM and Convolutional Neural Networks (CNN) that interprets sub-sequences of input data and extracts features that are provided as a sequence to the LSTM model. The methodology is validated using data sets from the Romanian Market Operator (OPCOM) and other market operators from Serbia (SEEPEX), Hungary (HUPX) and Bulgaria (IBEX). Our models improve the results for the day-ahead forecast in comparison with other models by 21.02% in terms of Mean Absolute Error (MAE).