In today's age, the dissemination of inaccurate and misleading facts on digital systems has ended up a serious problem that poses risks to people, organizations and society as a whole. Fake news spreads quicker than true information due to its bizarre and compelling content material. Fake information may have critical results, undermining the democratic system, undermining belief in the media and growing further divisions. Because of its fast spread and impact on the general public, it's far crucial to have a powerful manner to detect false information. In this paper, we pay attention to the usage of machine learning algorithms which includes SVM, Naive Bayes, and logistic regression, in addition to acquiring knowledge through deep learning models, which includes LSTM & Bi-LSTM, as well as utilising transfer learning with BERT-based approaches. Furthermore, the performance of these models will be evaluated on datasets comprising both Hindi and English languages.
Large language models (LLMs) are evaluated as if quality were a property of the model. We show that it is a property of the human–model system. In a benchmark of ten LLMs across 12 tasks with controlled prompt variation and a user study (n=80), we find that query context and user expertise jointly modulate response quality. Specifically, context enrichment improves BERTScore F1 by 0.07 for novice users but only 0.02 for experts, and model rankings shift by up to five positions across user-group conditions. More critically, we identify a fluency–accuracy dissociation: 62.5% of users prefer a factually incorrect answer when it is more fluent, and this preference is strongest among novice users who also provide the least context. Because RLHF-style alignment trains on these preferences, the human–model loop can converge on a stable equilibrium in which users are satisfied but answers are wrong. We formalise this as ΔQ ∝ ΔH(C) · σ(U), where σ(U) is the user’s prompt under specification, and verify that the model correctly predicts the monotonic ordering of context effects across user groups (3.5× ratio between novice and expert gains). We further show that the same context–prompt–receiver dynamics arise in live musical performance, where conductor gestures, orchestral generation, and heterogeneous audience reception exhibit structurally isomorphic behaviour. These results imply that single-score leaderboards are misleading, that current alignment methods may systematically optimise for perceived quality rather than factual correctness, and that evaluation frameworks should be explicitly user-parameterised.
The fast expansion of Internet of Things (IoT) infrastructures has made cybercrime more sophisticated, making traditional intrusion detection methods ineffective for detecting coordinated and changing attacks. Standard static feature learning or independent network-flow analysis methods offer poor explainability for forensic inquiry. This paper proposes EDGNN-Forensics, an Explainable Dynamic Graph Neural Network framework that models IoT communications as dynamic graphs and integrates graph representation learning, hybrid graph-based anomaly detection, explainable graph intelligence, and adaptive cybercrime decision optimisation into a single endto-end architecture to address these limitations The approach uses dynamic communication graphs to preserve temporal and structural relationships and graph attention and hybrid anomaly learning to identify known and unknown assaults. Node, edge, and analyst trust estimation are provided via an Explainable Graph Intelligence Layer for interpretable forensic analysis. Experimental evaluation using CICIoT2023 and NF-ToN-IoT benchmark datasets. The suggested model had 99.61% accuracy, 99.50% F1-score, and 0.999 AUC on the CICIoT2023 dataset and 99.42% accuracy, 99.27% F1-score, and 0.999 AUC on the NFToN-IoT dataset. This shows that EDGNN-Forensics outperforms graph-based and hybrid intrusion detection methods for accurate, explainable, and scalable cybercrime detection in current IoT systems.
This paper analyzes critically two graphical methods of representing sampled signals, which are exploited in the literature. Both are based on the description of the sampling operation with the use of a Dirac comb. However, they differ significantly from each other. It is explained here why this is the case and shown which of them is closer to reality. Reinterpretation of the first graphical representation discussed in view of the second one is also proposed. Finally, a remark about objects that are defined analytically but have no graphical representations is given.
This paper presents a methodology for optimizing the operation and sizing of a railway energy microgrid intended for island operation. The analyzed system integrates photovoltaic generation, battery energy storage, and traction substation load demand. The proposed model assumes that locally generated photovoltaic energy should primarily supply the traction load and charge the storage system, while energy export to the distribution grid is eliminated. A Particle Swarm Optimization algorithm is used to determine the required battery storage capacity under power balance, state-of-charge, depth-of-discharge and converter constraints. Simulations are performed for a traction substation using annual load data and photovoltaic generation profiles for south- and east-west-oriented panels. The results show a polynomial relationship between photovoltaic power and required storage capacity. For island operation, a one-hour autonomy criterion based on the P95 load level gives a storage capacity of 1437 kWh. State of Charge analysis confirms that this value covers most historical load windows, although extreme cases may still cause it to be violated.
