International Journal of Information and Communication Technology (13 papers in press)

Regular Issues

• Open Access

  Dynamic modelling of brand-user relationships via graph neural networks for enhanced marketing optimisation ( Free Full-text Access )

  by Aoxue Yao 
  Abstract: To address the limitations of existing methods in modelling dynamic and heterogeneous brand-user relationships, this study proposes a dynamic heterogeneous graph attention network (DHGAT). This framework integrates three core innovations: 1) a time-decay-based edge weighting mechanism that quantifies temporal dynamics of user-brand interactions; 2) a cross-relation attention layer that distinguishes semantic differences among diverse behaviours (e.g., purchases vs. complaints) through relation-specific gating; 3) a reinforcement learning decision engine optimising marketing actions via Q-learning. Validated on a real-world e-commerce dataset (32,000 users, 142M interactions), DHGAT achieves an AUC of 0.892 in relationship prediction (5.7%16.8% higher than baselines) and boosts marketing ROI by 41% in online A/B tests. The framework enables end-to-end optimisation of marketing strategies while balancing short-term conversions and long-term user value, offering a novel paradigm for data-driven marketing decision systems.
  Keywords: DHGAT; time-decay edge weighting; cross-relation attention; brand-user relationship modelling.
  DOI: 10.1504/IJICT.2025.10074024
   
  

• Open Access

  Enhancing cross-border e-commerce English text classification using graph neural networks and transfer learning ( Free Full-text Access )

  by Peng Du, Yanping Jiang 
  Abstract: Cross-border e-commerce platforms frequently feature English text characterised by mixed terminologies and informal syntactic structures, posing significant challenges for conventional classification models due to sparse labelled data. To address these limitations, this study introduces a novel classification framework that synergistically integrates GNNs and transfer learning. Specifically, a heterogeneous text graph incorporating word-document relationships is constructed to capture semantic dependencies, followed by the implementation of a domain-adaptive transfer mechanism to mitigate data sparsity through knowledge migration from related domains. Experimental evaluations on publicly available datasets, including Amazon Review and AliExpress, demonstrate that the proposed method achieves an accuracy of 92.7%, outperforming the BERT baseline by 4.5 percentage points. Furthermore, it significantly enhances classification efficacy in critical scenarios such as marketing content analysis and post-sale complaint resolution. This research advances cross-domain e-commerce text analytics by providing robust solutions for data-scarce environments.
  Keywords: graph neural networks; GNNs; migration learning; cross-border e-commerce; text categorisation; domain adaptation.
  DOI: 10.1504/IJICT.2025.10074257
   
  

• Open Access

  Style transfer of ink wash painting based on deep convolutional neural network and feature scaling ( Free Full-text Access )

  by Jia Zeng 
  Abstract: This study proposes AdaFSNet, a neural style transfer model tailored for mural images with intricate textures and cultural motifs. Leveraging adaptive feature scaling, AdaFSNet enhances style-content fusion while maintaining chromatic and structural integrity. Trained on 200 diverse samples from WikiArt, InkWash, and MuralSet, the model is evaluated on both seen and unseen mural styles. It achieves a PSNR of 25.8, SSIM of 0.86, and LPIPS of 0.17, outperforming baseline models such as AdaIN and MSG-Net. AdaFSNet demonstrates strong generalisation in zero-shot settings, offering practical value for digital heritage conservation and stylisation of culturally significant artwork.
  Keywords: neural style transfer; NST; adaptive feature scaling; mural image stylisation; reversible decoder; texture and colour preservation; zero-shot generalisation; heritage digitisation; PSNR; SSIM; LPIPS.
  DOI: 10.1504/IJICT.2025.10074258
   
  

• Open Access

  Vocational schools in the use of information technology by educational profiles ( Free Full-text Access )

  by Fakete Duraku 
  Abstract: Information technology, or IT enablement in education, is becoming an essential part of the digital world, forcing us to adapt to the IT-enabled new normal. Despite this, as an educational assessment, this exam evaluates the strong application of technology in different school settings to determine its suitability. This study investigates Kosovo vocational schools from three areas health, engineering, and business on the use of IT tools in the teaching and learning going on. This study is based on the technology acceptance model (TAM) and data from 638 contributors, explores the types of ICT being used, barriers to the use of ICT, and opportunity for growth. The results show that while key stakeholders in education and the labour market see value in ICT - the actual educational delivery is constrained by poor infrastructure, lack of training, and insufficient funding for tools and resources.
  Keywords: vocational schools; information technology; IT; educational profiles; software applications.
  DOI: 10.1504/IJICT.2025.10074322
   
  

• Open Access

  Application of evolutionary neural model in path optimisation of simulation system ( Free Full-text Access )

  by Meng Xu, Cheng Xu, Luyao Pei, Bo Wang, Zhixuan Lv 
  Abstract: This research introduces an evolving neural model that combines learning from neural networks with evolutionary computation techniques to optimise paths in simulation systems. The suggested approach overcomes the shortcomings of traditional algorithms when faced with dynamic, complicated situations by merging structural and parametric optimisation. To improve flexibility, convergence speed, and generalisation performance, the model uses an actor-critic reinforcement learning scheme, evolutionary field optimisation, and neural architecture search. Analyses of experimental data show that the suggested method is more efficient, stable, and accurate than more conventional methods like Q-learning and DQN. Path planning, risk minimisation, and live system simulation are three areas where the results show evolving neural models could improve intelligent decision-making.
  Keywords: topics covered include digital twins; computational intelligence; evolutionary algorithms; EAs; evolutionary neural networks; path optimisation; simulation systems; and actor-critical models.
  DOI: 10.1504/IJICT.2025.10074323
   
  

• Open Access

  ST-HGCN-enhanced real-time compensation for industrial robot positioning errors ( Free Full-text Access )

  by Mingxiong Wu 
  Abstract: Industrial robots frequently exhibit degraded positioning accuracy under dynamic coupling effects and environmental perturbations. To mitigate this, we introduce a real-time compensation framework powered by a spatio-temporal hybrid graph convolutional network (ST-HGCN). The methodology constructs a unified model integrating spatial sensor dependency graphs with temporal error propagation chains, utilising high-precision ground truth from the European Robotics Challenge (EUROC) micro aerial vehicle (MAV) dataset combined with inertial measurement unit (IMU) data. Experimental validation demonstrates a 62.3% reduction in root mean square positioning error (RMSE) relative to conventional graph-convolutional long short-term memory (LSTM) networks during complex multi-axis trajectories, while sustaining compensation latency under 2 ms. This work establishes a novel data-driven paradigm for high-precision robotic control, with direct applicability to precision manufacturing and flexible assembly operations requiring micron-level accuracy.
  Keywords: ST-HGCNs; industrial robotics; localisation error compensation; real-time control; sensor fusion; inertial measurement unit; IMU.
  DOI: 10.1504/IJICT.2025.10074324
   
  

• Open Access

  Vocational education course recommendation based on neighbours under the construction of knowledge graph ( Free Full-text Access )

  by Hantian Wang 
  Abstract: Focusing on issues of incomplete user information and inaccurate recommendations in current vocational education course recommendation methods, this paper first constructs a knowledge graph (KG) for vocational education. On this basis, a novel negative sampling approach is employed to enhance the KG representation model TransH (EOTransH), and different weights are given to negative samples assigned different scores contribute to full model training. Then, a bipartite graph of courses and users is constructed, and KG embedding is built through the joint user entity neighbourhood information. Furthermore, higher-order connectivity information between users and courses is mined through attention-based propagation. Finally, an attention network is built in the output prediction layer to explore user preference features. Experimental outcome on the MOOCCourse and MOOCCube datasets indicate that the proposed approach improves F1 by at least 2.05%, 4.09%, effectively solving the problem of inaccurate recommendations.
  Keywords: vocational education course recommendation; knowledge graph; nearest neighbour method; TransH model; attention network.
  DOI: 10.1504/IJICT.2025.10074331
   
  

• Open Access

  An intelligent recommendation system for music therapy resources based on a knowledge graph ( Free Full-text Access )

  by Ya Ni, Yingjin Shan 
  Abstract: To address the limitation of insufficient flexibility in current music therapy recommendation systems for capturing critical inter-entity relationships, this paper first uses the pre-trained language model BERT to enrich representation vectors of entities and relationships, and integrates the user-music therapy resource interaction graph into the knowledge graph, extracting collaborative information. The graph attention mechanism is introduced, allowing the weight of each neighbour node to be dynamically adjusted according to its relationship with the target node. Finally, the score that the user gives to the music therapy resource is predicted by calculating the dot product between the user representation and the music therapy resource representation, and the top N music therapy resources are recommended. Experimental results show that the AUC of the proposed model is improved by 4.8521% compared to the baseline model, 21%, which can accurately recommend music therapy resources that match the users preferences.
  Keywords: music healing resource; intelligent recommender system; BERT model; knowledge graph; graph attention network.
  DOI: 10.1504/IJICT.2025.10074332
   
  

• Open Access

  A furniture design assistance method based on spatiotemporal graph neural networks and multi-objective optimisation ( Free Full-text Access )

  by Tiande Wei 
  Abstract: This paper suggests a furniture design assistance method (FD-STGMO) based on spatial-temporal graph neural network (ST-GNN) and multi-objective optimisation to help with the problem of balancing the structural complexity and multi-objective optimisation needs in the furniture design process. The technique first employs ST-GNN to find structural change features. Then, it uses multi-objective optimisation algorithms to come up with design solutions. Finally, it builds a collaborative end-to-end design support system. The performance comparison tests done on the simulation dataset all show that the new FD-STGMO method is better than the old one in four areas: structural stability (0.84), material utilisation (0.88), functional adaptability (0.80), and aesthetics score (0.85). The findings of the modular contribution analysis experiments show that FD-STGMO has good potential for use in engineering and business.
  Keywords: furniture design; spatial-temporal graph neural network; ST-GNN; multi-objective optimisation; intelligent aided design.
  DOI: 10.1504/IJICT.2025.10074333
   
  

• Open Access

  Optimisation of academic gap compensation strategy based on transfer learning ( Free Full-text Access )

  by Yongmei Fu 
  Abstract: Identifying and addressing students academic gaps are essential for delivering effective personalised learning experiences. In this study, we present a transfer learning model that combines transformer layers with convolutional modules to detect learning deficiencies and recommend targeted exercises. The model analyses student interaction data from an online homework platform, capturing patterns that indicate areas of misunderstanding. By integrating both global sequence modelling and local feature extraction, the system predicts performance outcomes with high accuracy. In experiments, the model achieved 87.5% accuracy and an AUC of 0.91, outperforming traditional approaches across multiple benchmarks. It also processes each student sequence in under 0.15 seconds, supporting its practical use in real-time learning environments. These results confirm the models capability for accurate prediction, reliable gap detection, personalised intervention, and practical deployment in adaptive learning systems.
  Keywords: academic gap detection; transfer learning; transformer; personalised learning.
  DOI: 10.1504/IJICT.2025.10074334
   
  

• Open Access

  Optimised scheduling of network teaching resource management based on improved genetic algorithm ( Free Full-text Access )

  by Jing Zhang 
  Abstract: Network teaching resources provide convenience for daily teaching. Intending to issues of low scheduling accuracy and long response time in management methods of network teaching resources, this paper first optimises the genetic algorithm (GA) based on adaptive neighbourhood and wolf swarm algorithm. First, through the optimisation of encoding and initial population, an adaptive crossover operation based on greedy algorithm is designed. The reversal operation and variable neighbourhood search algorithm are used to complete the mutation operation of the population. Then, a mathematical model of network teaching resource scheduling is established, and the improved GA is used to solve the mathematical model, thereby obtaining the list of network teaching resource information after optimisation scheduling. Experimental results show that the management scheduling accuracy of the proposed method is 98.45%, and the response time is 0.23s, providing a feasible solution for the intelligent scheduling of network teaching platforms.
  Keywords: network teaching resources; management optimisation scheduling; genetic algorithm; GA; adaptive neighbourhood; wolf swarm algorithm.
  DOI: 10.1504/IJICT.2025.10074335
   
  

• Open Access

  Research on artistic style recognition and image transfer method based on deep visual feature extraction ( Free Full-text Access )

  by Wei Wang, Hanchen Li, Rushana Sulaiman @ Abd Rahim, Peilei Cui 
  Abstract: This research presents a comprehensive approach to artistic style recognition and image style transfer using deep visual feature extraction techniques. To enhance the identification of fine art forms, the study employs a two-stage classification model that combines shallow and deep neural networks, utilising convolutional neural networks (CNNs), namely VGG16 and VGG19. A novel neural style transfer network is proposed, incorporating a coarse-to-fine methodology and whitening and colouring transformation (WCT) to preserve global content structures while effectively applying local stylistic elements. Extensive experiments on the Wiki Art and Pandora 18K datasets validate the models ability to enhance style classification accuracy, minimise restructuring loss, and reduce runtime. The outcomes show that the suggested approach greatly improves automated art analysis and digital creative apps while keeping high-resolution image integrity and successfully integrating the unique visual traits of artistic styles.
  Keywords: artistic style recognition; image style transfer; deep learning; convolutional neural networks; neural style transfer; whitening and colouring transformation; WCT; shallow neural network; SNN; feature extraction; image processing; coarse-to-fine stylisation; computer vision; machine learning; Wiki Art.
  DOI: 10.1504/IJICT.2025.10074385
   
  

• Open Access

  AI-enabled association rule mining with cloud platforms for rural digital finance services ( Free Full-text Access )

  by Yuanyuan Yue, Qianqian Wang 
  Abstract: This study presents an AI-enabled association rule mining framework integrated with cloud platforms to enhance rural digital finance services. The proposed system leverages scalable cloud infrastructure for large-scale transaction analysis, enabling efficient identification of patterns and relationships within rural financial data. Results demonstrate improved service personalisation, fraud detection, and strategic decision-making. Rural digital finance faces challenges in data management, transaction security, and service personalisation. Leveraging AI-driven association rule mining with cloud computing can bridge these gaps by enabling real-time, scalable analytics. Previous studies have explored cloud-based financial analytics and AI-driven transaction pattern discovery, Existing methods often lack adaptability for low-resource environments. Preprocessing, and AI-driven association rule mining. Data from rural financial institutions is processed to uncover actionable patterns for decision-making. Experiments using large-scale rural transaction datasets achieved high pattern discovery accuracy and reduced processing time. The system demonstrated scalability and robustness under varying data loads.
  Keywords: multimodal motion analysis; inertial motion sensors; recurrent neural networks; graph-based neural networks; action recognition; intelligent sports training.
  DOI: 10.1504/IJICT.2025.10074336