There is a growing need to understand how live streaming e-commerce influences consumers’ purchasing behavior. Perceived value, engagement, and live streaming quality are crucial components that utilized structural equation modeling (SEM) to examine the factors that influence purchase intentions. This study presents a methodology for analyzing the variables that influence live streaming e-commerce purchase decisions. The study uses SEM and Machine Learning algorithms like Bayesian model, Random Forest, XGBoost, KNN and SVM to assess the prediction. This paper uses two feature transformation methods (MinMax and Zscore) and two feature selection models (InfoGain and Correlation) to improve the prediction of purchase intention. This paper gathers questionnaire responses from 500 participants who have purchased goods through E-commerce Live Broadcast in China and validates the results using a SEM. The study provides a reliability and validity analysis for the suggested model using SEM analysis. The attributes of live broadcasts can elevate the perceived value and trustworthiness, as well as consumer impulsivity, hence increasing customers’ likelihood to purchase.
Generating structured reports from event images is a crucial task across domains such as journalism, security analysis, and event documentation. Manual report writing is labor-intensive and time-consuming, motivating the need for automation. This paper introduces “Vision Narrate,” a novel AI-driven architecture that uniquely integrates state-of-the-art techniques in image captioning, context segmentation, and report generation. Unlike existing approaches, our system innovates by combining a fine-tuned VGG16-based encoder–decoder model for detailed image captioning with a new context identification strategy that employs MPNet sentence embeddings and a reverse sigmoid-weighted cosine similarity method to dynamically group captions into event-specific segments. Finally, a generative AI model is used to compose coherent, structured reports from these segmented captions. Experimental results on a dataset of 900 event images from 58 events demonstrate that our method achieves competitive performance with BLEU-4 (0.35), CIDEr (0.97), Pk (0.23), and WindowDiff (0.19) scores compared to traditional text segmentation baselines, while human evaluation yields an average coherence score of 4.74/5 and a fluency score of 4.88/5. Our findings indicate that “Vision Narrate” offers a scalable and efficient solution for automated report generation by uniquely fusing advanced techniques, paving the way for future enhancements in dataset diversity, factual accuracy, and multimodal context segmentation.
This paper presents an autonomous Unmanned Aerial Vehicle (UAV) navigation framework designed to track a programmable square-pulse altitude profile in real time. The system integrates a Cube Orange Plus flight controller with a Raspberry Pi companion computer through the Micro Air Vehicle Link (MAVLink) protocol, enabling dynamic switching between discrete altitude levels during GPS-guided missions. Unlike existing UAV systems that primarily emphasize continuous trajectory following, dense mapping, or learning-based altitude regulation, the proposed approach focuses on real-time discrete altitude modulation supported by a lightweight sensor-fusion architecture. A LiDAR-barometer fusion mechanism provides continuous altitude correction, ensuring stable climbs, descents, and sustained low-altitude cruise phases without the need for computationally intensive processing. Experimental flight results demonstrate reliable square-wave altitude tracking and robust horizontal stability under varied operating conditions, underscoring the practicality and efficiency of the system for structured and energy-aware aerial missions.
Smart environmental monitoring in urban areas demands highly efficient sensor networks capable of minimizing energy consumption while maintaining reliable 5G connectivity. This paper proposes a novel three-layer framework for energy-efficient 5G-enabled sensor networks tailored for sustainable cities. The Glass Sponge Topology Layer reduces signaling overhead and coverage redundancy by implementing a lightweight, lattice-inspired sensor deployment. The Electric Eel Energy Layer makes energy storage and transmission scheduling pulse-based to reduce idle energy consumption in the network. The Ant Colony Routing Hybrid Layer adaptively optimizes routing paths with pheromone-inspired scoring for low-latency and energy-efficient routing. Simulation tests on an urban deployment of 1 km² sensors demonstrate that the framework saves 32
Customer segmentation refers to the process of categorising customers into clusters that exhibit identical behaviour, enabling more effective product promotions and marketing strategies. The customer segmentation aims to investigate how to address customers in various forms. It helps companies recognise their valuable customers and fulfil their requirements by enhancing products and services and includes aspects such as demographic, geographic, psychographic, and behavioural. Recently, machine learning (ML) approaches have been used for customer and classification models that identify complex data patterns. This study presents an Advanced Machine Learning with Feature Selection for Robust Customer Segmentation with Purchase Behaviour Prediction (AMLFS-RCSPBP) approach. Initially, data pre-processing is performed to transform customer data into a usable format. Additionally, the information gain (IG) technique is used to select an effective set of features, and fuzzy-c-means (FCM) clustering is employed for segmentation. To estimate the optimum number of clusters, the Silhouette score is applied. Moreover, a multilayer perceptron (MLP) classifier is used to allocate the clusters to the unseen customers. Finally, the enhanced dung beetle optimisation (EDBO) technique is used to determine the optimal choice of MLP parameters. The comparison study of the AMLFS-RCSPBP method demonstrated a superior accuracy value of 94.75
Smart Grids incorporate advanced metering infrastructure and communication technologies to facilitate real-time monitoring and efficient control of power distribution systems. However, their large-scale deployment and open architecture introduce significant security concerns, particularly regarding device authentication and data integrity. Moreover, the advent of quantum computing poses a substantial threat to conventional cryptographic methods, necessitating the development of quantum-resistant security solutions. In this study, we present a novel authentication protocol for SGs that utilizes blockchain technology to establish decentralized trust and incorporates lightweight cryptographic primitives to enhance security. The proposed scheme is rigorously evaluated through formal verification methods, informal security analysis, and validation using the Scyther tool, affirming its robustness. Furthermore, performance analysis indicates that the scheme achieves a favourable balance between security and efficiency, demonstrating its suitability for deployment in resource-constrained smart grid environments.
A contemporary approach to address the problems brought on by urban transportation is the evaluation of sustainable mobility. In metropolitan areas, electric vehicles (EVs) specifically can improve the sustainability of transportation. This research uses a Knowledge-based Artificial Network (KANM) approach to create an efficient user behavioral framework focused on the Behavioral Learning Theory (BLT) to investigate Kerala customers’ perceptions of the shift to electric automobiles. To illustrate the results, information has been gathered from publicly accessible sources and calculated using finite element design and complicated variable connection analysis. According to the research findings, consumers in metropolitan areas plan to transition to electric vehicles according to subjective and attitude standards, such as the ratio of EV sales, environmental effects, barriers, and the cost of switching to EVs. The current Distributed Optimization Algorithm (DOA), which aims to give multi-objective restrictions, is used to examine the parameters and arrive at a solution in the Kerala region. The moderating impact demonstrates that the suggested approach performs better than the current methods in calculating the switching cost and creates an enhanced trade-off.
“Threshold quantum secret sharing" (TQSS) schemes of the form (q, n) provide enhanced practicality compared to traditional (n, n) models, offering improved scalability, fault tolerance, and operational flexibility. In this study, we propose a (q, n)-TQSS protocol utilizing “single photons" and “unitary phase shift" operations. The scheme leverages symmetric multivariate polynomials along with polynomial interpolation techniques to securely distribute both “classical information" and “quantum states". This method ensures the confidentiality and integrity of the transmitted data, while also incorporating mutual identity authentication between the dealer and participants to safeguard the reconstruction process. The original secret is accurately recovered through Lagrange interpolation, enabling precise reconstruction. Security analysis confirms that the proposed protocol is robust against a variety of adversarial strategies, including standard eavesdropping attempts and insider attacks such as “entanglement swapping". Furthermore, the scheme is characterized by its simplicity, ease of implementation within physical systems, and adaptability to a wide range of practical applications, making it a highly efficient and viable alternative to existing quantum secret sharing (QSS) protocols.
Employee digital strain has emerged as a critical factor affecting workplace productivity and well-being especially in the modern fast paced work environment. Accurately detecting overwork, multitasking, and burnout requires modeling complex behavioral and relational dependencies from multimodal digital activity data. In this work, we propose AutoEG, an automated graph-based contrastive learning framework for robust employee stress representation and classification. Layer 1 comprises multimodal behavioral encoder (MBE) which extracts embeddings from keystrokes, application usage logs, and chat/email sentiment and wearable physiological signals to represent individual digital behavior. Layer 2 has temporal graph neural network (TGNN) with graph Laplacian Regularization for capturing intra- and inter-employee dependencies across tasks, tools, and interactions while preserving temporal and relational structures. Layer 3 contains Self-Supervised Contrastive Learning Module which leverages latent embeddings to automatically distill meaningful cross-employee patterns thus enhancing robustness against noise and distribution heterogeneity. Layer 4 is the Fusion and Classification Layer which integrates embeddings from all prior layers to generate risk scores for digital strain, overwork, and techno-stress. Experimental evaluation on multiple workplace datasets demonstrates that AutoEG achieves the highest digital strain accuracy (DSRS) thus outperforming baseline machine learning and graph neural network models.
Customer segmentation involves clustering customer data by shared behaviours, which can assist businesses anticipate purchasing behaviour, identify potential customers, and create targeted marketing campaigns. Business organizations can improve resource allocation by comprehending diverse customer group. Deep learning (DL) models ease segmentation by discovering hidden patterns in intrinsic datasets. Techniques like deep clustering, autoencoders (AE), and neural networks capture nonlinear relationships in customer data for more accurate and dynamic segmentation. In this study, an Enhanced Sand Cat Swarm Optimisation with Deep Learning Assisted Customer Segmentation with Purchase Behaviour Analytics (ESCSODL-CSPBA) technique is proposed. Initially, the Recency, Frequency, Monetary, and Repurchasing Number of Times (RFM-RN) model is utilized. Additionally, ESCSO and DBSCAN models are employed for feature subset selection and clustering. Finally, the Convolutional Recurrent Neural Network (CRNN) technique is implemented for classification. The comparison study of the ESCSODL-CSPBA approach depicted superior accuracy of 95.90
