International Journal of Information and Communication Technology (12 papers in press)

Regular Issues

• Open Access

  Application research of a LightGBM-based predictive evaluation model for college English learning behaviour ( Free Full-text Access )

  by Xinxin Yang 
  Abstract: In order to meet the demand for the digital modelling and intelligent analysis of students learning behaviour, a predictive evaluation framework is proposed, which combines feature engineering with LightGBM. Through the construction of tags and the extraction of training samples from college English course behaviour log, the construction strategy and interpretable output mechanism of the model are discussed. The results show that the proposed model achieves an F1-score of 0.889 on the test set, with an average response latency of less than 230 ms, and a performance fluctuation of less than 1.6% in 5 times cross validation, demonstrating robust deployment stability and generalisation.
  Keywords: learning behaviour prediction; LightGBM algorithm; educational data mining; Intelligent teaching analytics; learning behaviour modelling.
  DOI: 10.1504/IJICT.2027.10080149
   
  

• Open Access

  Comprehensive evaluation of key indexes of power engineering based on K-nearest neighbour algorithm ( Free Full-text Access )

  by Yulong Li, Wei Ni, Bo Chen 
  Abstract: To improve the accuracy of key indicator evaluation in power engineering, this study proposes a comprehensive evaluation method based on K-nearest neighbour (KNN) algorithm. Aiming at the limitations of traditional methods in processing nonlinear and high-dimensional data, an evaluation system including ten key indicators such as engineering cost, construction period, and equipment performance was constructed to conduct experiments on 50 power engineering project data in a certain region. The results show that compared with traditional methods, the KNN model has a 15% improvement in evaluation accuracy, a more reasonable allocation of indicator weights, and can effectively identify key factors affecting engineering quality and efficiency. It also exhibits strong robustness in dealing with nonlinear problems and is suitable for power engineering of different scales and types. This method provides a scientific basis for decision-making in power engineering and has promotional value.
  Keywords: KNN algorithm; power engineering; index analysis; comprehensive evaluation.
  DOI: 10.1504/IJICT.2026.10080314
   
  

• Open Access

  Automatic graphic-text composition for graphic design using diffusion models and layout optimisation ( Free Full-text Access )

  by Ju Ling 
  Abstract: In graphic design practice, traditional manual iteration is time-consuming, while existing generative artificial intelligence tools often fail to accurately capture design semantics, layout rules, and user intentions, leading to outputs that deviate from professional requirements. This paper presents a graphic-text synthesis model based on diffusion models and layout optimisation. The model consists of three parts: an intent encoder that transforms natural language into design constraints, a layout generator with spatial optimisation to prevent overlap and boundary violations, and a diffusion-based style transfer module for consistent content filling. Results on the public Peking University layout dataset and Crello dataset show that the proposed model outperforms baselines, achieving an 8.2% improvement in F1-score and an area under the curve of 0.93. All metric improvements are statistically significant (p < 0.01). This approach reduces design iterations and offers a new direction for human-machine collaborative creation.
  Keywords: generative AI; graphic design; diffusion model; layout optimisation; text-image synthesis.
  DOI: 10.1504/IJICT.2026.10080869
   
  

• Open Access

  UrbanGreen-LiteFormer: a lightweight vision model for disease and pest recognition of urban landscape plants in complex urban green-space environments ( Free Full-text Access )

  by Na Li, Xianxian Yu 
  Abstract: The visual context of landscape plant management in urban settings is complex and not like controlled crop datasets. Leaves are often found near roads, buildings, walls, soil, shadows, vehicles, people and other vegetation, and small areas of the image are filled with disease spots and pests. Moreover, healthy leaves are polytypic, which induce a degradation in the performance of the conventional CNN, and complete Transformer models are prohibitively expensive to deploy in the field. This paper proposes UrbanGreen-LiteFormer, a lightweight hybrid vision model for disease and pest recognition in urban green spaces. The ULPDP-12640 dataset contains 12,640 images collected from 42 urban scenes in Nanjing, China, covering 18 disease, pest, and healthy categories. The model integrates an urban texture stem, multi-scale pest-disease block, background-suppressed efficient attention, and edge-calibrated channel mixer. Across five runs, it achieved 92.41% accuracy and 92.05% macro-F1 with 4.26 million parameters and 0.93 GFLOPs, supporting efficient Jetson Nano deployment with calibrated INT8 quantisation for urban pest control.
  Keywords: urban landscape plants; plant disease recognition; pest recognition; lightweight deep learning; hybrid vision model; edge deployment.
  DOI: 10.1504/IJICT.2026.10080898
   
  

• Open Access

  A secure and scalable blockchain framework for enhancing transaction transparency and operational efficiency in banking systems ( Free Full-text Access )

  by Yuan Yuan, Cheng Ju 
  Abstract: The banking industry is going digital very quickly, so it needs frameworks that make sure data is safe, operations run smoothly, and transactions are managed in a clear way. Blockchain technology offers decentralised trust, but combining it with traditional banking systems is hard because of issues like scalability, interoperability, and privacy. Traditional banking systems depend on centralised transaction verification, which raises the chances of single-point failure, data tampering, and delays in processing. Limited real-time auditing and traceability make it even harder to follow the rules and make customers feel safe. This study suggests the Hybrid Consensus driven adaptive blockchain framework (HCABF) as a way to make banking systems safer. The framework combines delegated proof of trust with adaptive byzantine fault tolerance to find a balance between security and scalability. HCABF achieves 23% higher throughput, 22%28% lower latency, 91%94% consensus efficiency and 94% fraud detection accuracy than CAM, CEM and BAM.
  Keywords: blockchain; banking systems; hybrid consensus; transparency; scalability; security.
  DOI: 10.1504/IJICT.2026.10080926
   
  

• Open Access

  Deep learning-driven modelling of acoustic logging signals and identification of complex reservoir fractures ( Free Full-text Access )

  by Jingping Wang, Quanxi Bao 
  Abstract: To address the challenges in fracture identification within complex reservoirs, including traditional methods inefficiency and dependence on precise parameters, and conventional models need for extensive labelled data with limited generalisation, this paper proposes an integrated deep learning method for acoustic logging signal modelling and identification. By constructing a conditional generative adversarial network, high-fidelity acoustic logging signals are efficiently synthesised, achieving a correlation coefficient of 0.92 between simulated and real waveforms and reducing the root mean square error by 30%. Using this generated data for training enhancement, the fracture identification model reaches 94.5% accuracy and 89.2% recall on the test set, significantly outperforming the unaugmented baseline. This approach reduces reliance on scarce and costly labelled data in well logging interpretation and provides a new tool for intelligent reservoir evaluation.
  Keywords: acoustic logging; fracture identification; data augmentation; complex reservoirs.
  DOI: 10.1504/IJICT.2026.10080927
   
  

• Open Access

  Financial risk monitoring based on corporate ESG performance from the perspective of sustainable development ( Free Full-text Access )

  by Jianmei He, Lihua Bi 
  Abstract: This study constructs a multi-source ESG rating system via entropy weight-TOPSIS and embeds ESG factors into a modified Z-score model. Using Chinese A-share data from 2015 to 2024, a 0.1 ESG score increase raises Z-score by 0.238, confirming risk mitigation, especially for non-state-owned and small-scale, high-competition firms where prediction errors decline by over 29%. To capture inter-firm risk transmission, an explicit-implicit hybrid association graph neural network integrates observable supply chain and equity linkages with attention-based latent correlations. It achieves the lowest MAE (0.81) and RMSE (2.07) against XGBoost, random forest, and transformer, with Diebold-Mariano significance (p < 0.01) and a Brier score of 0.063 (0.058 after temperature scaling). This research provides a dynamic early-warning tool for sustainable financial risk monitoring and differentiated ESG strategy allocation.
  Keywords: sustainable development; financial risk warning; multi-source data fusion; explicit-implicit association network; dynamic risk monitoring.
  DOI: 10.1504/IJICT.2026.10080962
   
  

• Open Access

  Dynamic budget allocation via deep Q-networks with embedded financial risk control using VaR/CVaR metrics ( Free Full-text Access )

  by Yifei Guo 
  Abstract: In view of the lack of adaptability and refined risk control of the DQN-based budget allocation model across cycles and contexts, this paper proposes a dynamic intelligent financial decision-making model integrating VaR and CVaR. The state space consists of 12-dimensional dynamic variables, which include macroeconomic cycles and market risk indices; the action space defines departmental budget proportions (minimum step 5%) and adjustable risk reserves. The dual-channel reward embeds penalties for risk overflow and budget deviation, balancing economic benefits (0.45) and risk control (0.35). The model is trained on 320 sets of eight-year enterprise financial data using MSE loss and Adam optimiser (learning rate 0.001, discount factor 0.95), achieving allocation accuracy of 92.1%, 87.5%, and 82.3% in stable, mild, and severe periods, and controlling risk overflow at 3.2%, 8.7%, and 14.5% respectively, outperforming traditional DQN. Cross-cycle adaptive budget optimisation is enabled as the loss converges to 0.08 within 200 rounds.
  Keywords: deep Q-network; dynamic budget allocation; financial risk control; intelligent decision-making; cross-cycle adaptation.
  DOI: 10.1504/IJICT.2026.10080963
   
  

• Open Access

  Construction of TPACK knowledge integration profile for university teachers based on generative AI and multimodal data ( Free Full-text Access )

  by Shaoxiong Tan 
  Abstract: Traditional assessment methods relying on scales or single observations are difficult to accurately depict the true teaching level of teachers in complex classroom situations. Therefore, based on generative AI and multimodal data, a method for constructing TPACK knowledge integration portraits of university teachers was studied. A three-dimensional portrait model covering four dimensions of knowledge, behaviour, ability, and development was designed based on the TPACK theory, aiming to systematically depict the cognitive structure, practical performance, decision-making literacy, and growth trajectory of teachers. By extracting behavioural features from multimodal data such as teaching videos, design texts, and platform logs, and quantifying and clustering them, a standardised evaluation input is constructed. On this basis, a hybrid evaluation model combining generative AI (big language model) and discriminative model (gradient boosting decision tree) is used to achieve objective and efficient quantitative scoring. Through deep semantic understanding, teaching behaviours are inferred and integrated portraits are generated. And it has been proven that it has advantages in quantitative consistency, situational matching, and growth trajectory fitting, effectively verifying its scientific, accurate, and developmental nature.
  Keywords: generative AI; multimodal data; university teachers; TPACK; portrait construction.
  DOI: 10.1504/IJICT.2026.10080973
   
  

• Open Access

  Robust multi-agent simulation modeling of the distribution of source-load asymmetric resource regulation capabilities ( Free Full-text Access )

  by Hongxuan Zhang, Jianxin Zhang, Qin Gao 
  Abstract: Under a high proportion of grid-connected renewable energy, source-side photovoltaic/wind power fluctuates randomly while load-side flexible loads and storage exhibit heterogeneous response characteristics, making system regulation assessment face the dilemma of unknown distributions. This paper integrates Wasserstein distributionally robust optimisation with multi-agent simulation to construct a collaborative modelling framework for source-load heterogeneous resources, filling the gap between overly conservative robust optimisation and distribution-dependent stochastic programming. Experiments on public datasets show that distributed robust optimisation-driven multi-agent simulation reduces the regulation power mismatch rate from 18.7% to 9.3%, increases the confidence self-regulation elasticity coefficient by 42% (from 0.38 to 0.54), and maintains schedulable potential coverage deviation within 4.1% even when the sample size is only 30% of the full distribution. This provides a quantifiable simulation tool for scheduling heterogeneous resources under high uncertainty.
  Keywords: distributionally robust optimisation; multi-agent simulation; heterogeneous resources; Wasserstein ambiguity set; regulation capability.
  DOI: 10.1504/IJICT.2026.10080991
   
  

• Open Access

  GAN-based English metaphor identification and paraphrase generation via optimised compression ( Free Full-text Access )

  by Xi Wu 
  Abstract: This study proposes an English metaphor identification and paraphrase generation model that integrates a generative adversarial network with an optimised compression algorithm to address contextual dependence and semantic ambiguity. The model employs adversarial training to enhance metaphor representation, generation, and discrimination, combined with a bidirectional transformer encoder for deep semantic encoding, and BiLSTM with attention for feature fusion. Principal component analysis and extreme learning machine are further integrated to improve training and inference efficiency. Experiments on the VU Amsterdam metaphor dataset and metaphor usage finder dataset demonstrate a recognition accuracy of 92.03%, latency of 13.2 ms, and 97.11% agreement with human judgments. On the metaphor understanding challenge and contemporary news headlines datasets, the model achieves a lexical overlap ratio of 0.987 with a generation time of 100.4 ms, demonstrating high accuracy, efficiency, and paraphrase quality.
  Keywords: generative adversarial network; GAN; optimised compression algorithm; metaphor identification; paraphrase generation; principal component analysis; PCA.
  DOI: 10.1504/IJICT.2026.10080992
   
  

• Open Access

  Comparative computation of multimodal semantic representations for ideological and value orientations ( Free Full-text Access )

  by Yanxia Wang, Ke Xue 
  Abstract: Currently, when multimodal models handle high-level semantic concepts such as ideological literacy value orientation, they often encounter the problem of fine value features being lost due to modality gaps, and are unable to effectively distinguish between factual statements and value judgements. This paper proposes a multimodal semantic representation method called literacy-aligned multimodal semantic representation based on difference-aware contrastive learning. By dynamically adjusting the contrast intensity of the text-image pairs and combining hierarchical regularisation, it constructs a progressive representation structure from the knowledge layer to the value layer in the shared space. Experiments on two public educational multimodal benchmarks show that this method improves the accuracy of text-image matching from 75.8% to 83.2%, and reduces the modality gap ratio from 1.02 to 0.87, significantly enhancing the discriminative ability for cross-modal value orientation.
  Keywords: multimodal representation; contrastive learning; value orientation; semantic alignment.
  DOI: 10.1504/IJICT.2026.10080993