Forthcoming Articles
International Journal of Information and Communication Technology
Forthcoming articles have been peer-reviewed and accepted for publication but are pending final changes, are not yet published and may not appear here in their final order of publication until they are assigned to issues. Therefore, the content conforms to our standards but the presentation (e.g. typesetting and proof-reading) is not necessarily up to the Inderscience standard. Additionally, titles, authors, abstracts and keywords may change before publication. Articles will not be published until the final proofs are validated by their authors.
Forthcoming articles must be purchased for the purposes of research, teaching and private study only. These articles can be cited using the expression "in press". For example: Smith, J. (in press). Article Title. Journal Title.
Articles marked with this shopping trolley icon are available for purchase - click on the icon to send an email request to purchase.
Online First articles are also listed here. Online First articles are fully citeable, complete with a DOI. They can be cited, read, and downloaded. Online First articles are published as Open Access (OA) articles to make the latest research available as early as possible.
Articles marked with this Open Access icon are Online First articles. They are freely available and openly accessible to all without any restriction except the ones stated in their respective CC licenses.Register for our alerting service, which notifies you by email when new issues are published online.
International Journal of Information and Communication Technology (36 papers in press) Regular Issues
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
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 An international communication influence model for intangible cultural heritage using graph transformers and MoE architecture by Jing Zeng Abstract: International dissemination of intangible cultural heritage (ICH) is difficult to assess because relevant data are heterogeneous, relational, and dynamically evolving. We propose GT-MoE-NIA, an influence assessment model that combines a graph transformer with a mixture-of-experts (MoE) architecture. A heterogeneous graph represents ICH projects, countries or regions, media, audiences, and their interactions. The graph transformer captures global structural and semantic dependencies, while a gating network dynamically routes representations to specialised experts for influence prediction. We evaluate the model on the Global Non-Material Cultural Heritage Propagation Dataset (GNCPD), integrating UNESCO inventories, GDELT news, Twitter data, and Wikipedia page views. Compared with GCN, GAT, and GraphSAGE baselines, GT-MoE-NIA reduces root mean square error by 18.7% and improves NDCG@10 by 22.4%. Expert-activation patterns also provide interpretable evidence of how dissemination modes affect influence, supporting data-driven resource allocation for international cultural promotion. Keywords: intangible cultural heritage; ICH; international dissemination; influence model; graph transformer; mixture of experts; heterogeneous graph neural network.  This article is being considered for Open Access
Abstract: Traditional intelligent English teaching platforms overly rely on students behavioural data, which often leads to reduced recommendation accuracy under data sparsity conditions. To address this issue, this study develops an intelligent English teaching platform that integrates an improved collaborative filtering algorithm with fuzzy C-means clustering, enabling multi-dimensional feature modelling and personalised recommendation. Experimental results show that the proposed method outperforms baseline algorithms in terms of convergence speed and recommendation stability, with precision reaching 0.76, recall stabilising at 0.74, and an F1-score of 0.86. On a self-constructed dataset, the model further improves recommendation accuracy and learning matching performance, maintaining high precision and recall, as well as stable hit rate and learning adaptability scores. Overall, the proposed platform effectively enhances recommendation performance and personalisation capability, providing a new technical approach for intelligent English teaching. Keywords: fuzzy c-means clustering; collaborative filtering algorithm; fuzzy system integration; FSI; English learning; intelligent teaching platform. DOI: 10.1504/IJICT.2026.10081218
Abstract: With the ongoing advancement of artificial intelligence technologies, personalised learning path design has become a critical focus for enhancing learning outcomes. This study presents a Transformer-based approach to English vocabulary learning path planning, aimed at dynamically adjusting learners study trajectories to improve both learning efficiency and memory retention. Compared with traditional recommendation algorithms-such as collaborative filtering, K-nearest neighbours, and content-based methods-experimental results show that the Transformer model performs superiorly in terms of path coherence, learning efficiency, and retention. This model demonstrates particular strength in personalised path adaptation. Among enhancing learners, accuracy improved significantly from 83.8% to 86.4% between rounds 20 and 30. For Level 1 (beginner) learners, the proposed method achieved a learning efficiency of 2.8 words per minute, marking a 12% increase over the best performance of traditional methods (2.5 words per minute via matrix factorisation). Keywords: transformer model; English vocabulary learning; learning path planning; personalised recommendation; self-attention mechanism. DOI: 10.1504/IJICT.2026.10081219
Abstract: The internet communication has multi-modal interaction in the symbolic exchange, media affordances, and dynamics of users. Existing techniques are challenged by symbolic imbalance, the disproportionate dominance or misrepresentation of communicative symbols across channels, and language inflation, where repetitive or inflated content diminishes interpretability and contextual coherence. To overcome these limitations, we propose the multimedia contextual and symbolic communication analysis framework (MC-SCAF) based on media context theory and structured multimodal feature analysis. MC-SCAF detects differences in symbolisation and linguistic meaning across channels using textual, visual, and interactive indices. Techniques such as context weighting, symbol density mapping and inflation indices are used to detect communication distortions and normalisation and contextual alignment are employed to reduce symbol dominance, redundancy and semantic drift. The experimental results show high accuracy of 91.3% for symbolic imbalance detection and 88.7% for language inflation evaluation, indicating that MC-SCAF produces clear and interpretable insights into multimodal communication patterns. Keywords: multimedia contextual and symbolic communication analysis framework; MC-SCAF; contextual communication; symbolic imbalance; language inflation; multimodal analysis. DOI: 10.1504/IJICT.2026.10081220
Abstract: Globally, academic researchers have found the convenience of using AIGC technology. It has turned into a huge or major transformation and innovation. This technology has been widely explored and applied in research. It gives an avenue for many teachers to transfer literacy in many universities. Known as a disruptive technology, AIGC creates far-reaching effect on pedagogical practices of university teachers more specifically in the deployment approach of data literacy instruction. There is an assumption that the reception of this kind of technology explicitly resolves its operational outcome and its extent of integration. This investigation intends to assess college educators contemporary context on the acceptance of AIGC guided by the theory of acceptance and use of technology (UTAUT) model. In this data-driven research, essential determinants influencing teachers application of AIGC technology to information literacy instruction are identified. Ultimately, the study calls for tailored implementation strategies and targeted guidance based on educators skill levels, experience, and instructional needs. It recommends to promote the deep integration of AIGC technology and college information literacy education, enrich the application scenarios of UTAUT framework in the field of educational technology, and provide theoretical support and practical reference for the high-quality development of college information literacy education. Keywords: information literacy; generative artificial intelligence; AIGC; UTAUT; technology acceptance; influencing factors. DOI: 10.1504/IJICT.2026.10081221
Abstract: This study proposes an end-to-end joint knowledge extraction model (E2E-JKEM) for history, addressing cognitive construction in history teaching, ancient-modern semantic gaps and complex relation overlaps in ancient Chinese auto-processing. The study creatively blends the cascade architecture of cascade relative triple extraction (CasRel) with the Siku Quanshu-a robustly optimised BERT pretraining approach (Siku-RoBERTa) pre-training model. The China Biographical Database Project (CBDB) for distant supervision to construct a high-quality corpus, extracting historical knowledge accurately from unstructured classics like Twenty-Four Histories (a collection of official historiographical works in ancient China). Experimental results show the model scores 4.75 points (out of 5) in the manual evaluation of teaching effectiveness and an 87.53% F1 value in the task of historical relationship extraction, which is significantly better than the conventional BERT-BiLSTM-CRF model. This study verifies that the generated knowledge graph (KG) clarifies historical context and deepens historical understanding, serving well for humanities research. Keywords: knowledge graph; Siku Quanshu-a robustly optimised BERT pretraining approach; Siku-RoBERTa; cascade relative triple extraction; CasRel; relationship extraction; wisdom education. DOI: 10.1504/IJICT.2026.10081260
Abstract: Classical anomaly detectors for Kaplan turbine friction monitoring operate on single SCADA windows, are hard to interpret, and register a friction transition only after onset. This paper proposes the hydraulic physics-guided contrastive temporal change-point network (HPG-CTCN), which detects incipient friction changes from existing SCADA measurements. It constructs 13 hydraulic physics-guided features, learns temporally discriminative representations through a self-supervised contrastive encoder with efficient channel attention (ECA), and reformulates detection as a focal-loss three-class problem over pre-transition, transition, and post-transition windows. The transition-class probability yields a risk score for early-warning alarms. On two full-scale Kaplan units, HPG-CTCN attained macro-F1 of 0.89 and 0.88 and transition-class F1 of 0.85 and 0.84, and, uniquely among the evaluated methods, raised alarms 4.2 h and 4.8 h before the transitions rather than after onset. Integrated gradients, SHAP, and attention identified pressure-difference and servo-pressure channels as the dominant predictors. Results are reported as proof-of-concept, not fleet-wide validation. Keywords: Kaplan turbine; hydropower SCADA; industrial internet of things; IIoT; friction monitoring; change-point detection; contrastive learning; physics-guided learning; early-warning analytics. DOI: 10.1504/IJICT.2026.10081261
Abstract: Transmission line galloping constitutes one of the most severe disasters threatening the safe operation of power grids during winter. Based on authentic galloping event observations, disaster loss records, and engineering failure analysis reports provided by Hubei power grid, this paper systematically investigates the spatial-temporal distribution characteristics, structural failure mechanisms, and core influencing factors of transmission line galloping. To ensure scientific rigor, a systematic methodology incorporating data quality control and statistical validation frameworks was established. The research results show that galloping disasters are primarily concentrated in the Jianghan Plain region, driven by the coupling effect of freezing rain and canyon wind channels. Quantitative forensic evidence reveals that 97.2% of severe equipment damage occurs in spans exceeding 300 metres, and longitudinal unbalanced dynamic loads induced by asymmetric tension sections (ratio >2.5) are the primary triggers for tower collapse. Furthermore, existing anti-galloping measures exhibit distinct limitations under extreme meteorological conditions. The conclusions of this study provide critical theoretical guidance and a quantitative foundation for disaster prediction, structural reinforcement, and intelligent early warning of smart grids. Keywords: transmission lines; galloping; failure mechanism; statistical validation; smart grid early warning. DOI: 10.1504/IJICT.2026.10081266
Abstract: Drowning represents a critical global public health issue, ranking as the third leading cause of unintentional injury mortality worldwide. Incidents predominantly occur in natural water settings, characterised by their sudden and unpredictable nature. Furthermore, extreme weather conditions, complex topography, and potential secondary accidents significantly exacerbate rescue risks and operational challenges. Therefore, developing a robust automated drowning detection system is urgently needed. Leveraging convolutional neural networks (CNNs) for object detection in complex visual scenes and considering the deployment constraints of unmanned surface vessels (USVs), this paper proposes an improved YOLOv8 model. Specifically, ShuffleNetV2 is adopted as a lightweight backbone, frequency channel attention network (FcaNet) is embedded to suppress glare and wave interference, and Wise-IoU v3 (WIoUv3) is employed to improve localisation accuracy for low-contrast drowning targets. Experimental results on the NVIDIA Jetson Orin Nano platform show that the proposed method achieves mAP@0.5 of 82.5% and 32.4 FPS, meeting real-time requirements in complex aquatic environments. Keywords: Yolov8 optimisation; drowning detection; ShuffleNetV2; FcaNet; WIoUv3; convolutional neural networks; CNNs; unmanned surface vessels; USVs. DOI: 10.1504/IJICT.2026.10081267
Abstract: This study proposes a quantitative evaluation framework to evaluate the influence of traditional cultural elements on user preferences in game design. In order to solve the problem of limited subjectivity and repeatability of existing evaluation methods, a comprehensive evaluation model combining analytic hierarchy process with weighting is proposed. A multi-level index system covering content characteristics, design consistency, and gameplay integration was constructed to evaluate traditional elements from multiple perspectives. Experimental results showed that the proposed method achieved stable and superior performance across different evaluation dimensions. In particular, the model obtained preference scores of 0.781 for traditional visual element density, 0.768 for cultural symbol frequency index, and 0.792 for narrative content coverage ratio, outperforming comparison models. The results demonstrate that the proposed framework can effectively characterise the multi-dimensional design quality and potential preference attractiveness of traditional cultural elements in games. Keywords: traditional cultural elements; user preference evaluation; AHP; weighted comprehensive evaluation model; game design analysis. DOI: 10.1504/IJICT.2026.10081301
Abstract: Streaming recommendation systems often reduce music preferences to static labels, overlooking the dynamic role that emotional experiences play in the formation of preferences. Addressing the challenge that existing algorithms struggle to capture the emotion-style generation mechanism, this paper proposes an affective-driven preference generation framework that enables the system to derive style preferences in real time based on valence and arousal. On the database for emotional analysis in music dataset and a self-built streaming behavior dataset, the framework achieved an area under the curve of 0.892, representing a 6.1% improvement over the best baseline. Furthermore, removing the core affective generation module resulted in an 18.7% drop in accuracy, confirming that affective computing is not merely an auxiliary matching feature but a constructive condition that drives the generation of stylistic preferences. Keywords: affective computing; style preference generation; streaming recommendations; circular affect model; collaborative filtering. DOI: 10.1504/IJICT.2026.10081302
Abstract: AI-assisted learning environments have changed the game in educational technology, allowing for personalised feedback and adaptive instruction, particularly in difficult subjects such as chemistry. However, there are few studies to investigate academic achievement and metacognitive development together. This study introduces Artificial Intelligence Mediated Chemistry Learning Model (AIM-Chem), a mixed-methods framework that incorporates adaptive feedback, predictive analytics, and metacognitive monitoring to foster conceptual understanding, self-regulation, and reflective learning. Experimental groups using AI-supported platforms were compared to traditional learning groups. Data were collected through standardised tests, metacognitive inventories and interviews. The results showed that learners using AIM-Chem had a better conceptual understanding, problem-solving accuracy and knowledge retention. Qualitative findings also indicated increased reflective thinking and self-regulated learning behaviours. The proposed framework achieved 82% learning performance, 95% interaction intensity, 4.7 metacognitive score, 5% error rate, 80% retention and 96% feedback utilisation. It confirmed the effectiveness of the proposed framework in AI-assisted chemistry education. Keywords: AI learning; chemistry education; metacognition; mixed methods; student performance. DOI: 10.1504/IJICT.2026.10081303
Abstract: This study proposes an advanced multimodal information visualisation framework for augmented reality (AR)-based cultural experiences to address the limitations of single visual modalities, cognitive overload, and lack of context-awareness in existing systems. The framework integrates multi-dimensional context perception (utilising a Gaussian-Newton optimisation posture estimation algorithm), a cross-modal self-attention data fusion network, and an adaptive mapping engine based on cognitive load theory (CLT). It successfully replaces traditional 2D interfaces with 3D embodied interactions. A controlled experiment (N = 120) in a museum setting revealed that this framework significantly increases average stay time and spatial exploration rates. Furthermore, it substantially reduces users' external cognitive load and operational frustration, providing solid theoretical support and an engineering paradigm for the next generation of intelligent digital cultural heritage systems. Keywords: augmented reality; multimodal information visualisation; embodied interaction; digital cultural heritage; cognitive load theory; CLT; multimodal data fusion; context awareness. DOI: 10.1504/IJICT.2026.10081304
Abstract: Personalised learning systems are being increasingly adopted in higher education institutions to enhance students academic outcomes. As there is a significant need for models that adapt to diverse learning behaviours and assessment patterns, this research utilises the collaborative filtering (CF) method, thereby enhancing individual learning experiences. A novel memory-enhanced sequence network (MESeq Net) is introduced to capture temporal dependencies in student learning behaviours. Assessment dataset includes exam scores, assignments, quizzes, participation logs, and LMS interaction histories. long short-term memory-based sequence-to-sequence (LSTM-Seq2Seq) model processes sequential learning events, encoding student activity patterns and decoding predicted performance trajectories. Python-based implementation delivers accurate prediction models, generating personalised recommendations (PRs) with high accuracy (0.98). Results demonstrate improved learning outcomes, increased engagement rates, and strong scalability for large datasets, validating the models efficiency in real academic environments. Keywords: personalised learning; memory-enhanced sequence network; MESeq Net; recommendation system; higher education. DOI: 10.1504/IJICT.2026.10081305
Abstract: Natural language processing (NLP) in finance is growing, yet classical translation algorithms struggle with grammatical adaptation and semantic deviations in complicated financial documents. This study proposes an adaptive Chinese-English translation optimisation framework combining Fama factor analysis and multi-level deep learning (DL) to address this problem. First, we build a dynamic syntactic tree parser based on the Fama factor model to extract syntactic features from financial texts accurately. Second, a hybrid neural network architecture is designed and adversarially trained on a large-scale parallel corpus to produce dynamic grammatical adjustment strategies for the target language. Experimental results on the Reuters financial news dataset indicate a 15.2% improvement in BLEU score and an 18.7% decrease in grammatical errors over the transformer model. Semantic coherence increases by 22% and inference time decreases by 19.8% and error propagation is reduced by 41% in specialised financial scenarios. Keywords: Fama factor model; grammar structure; deep learning; financial texts; automated optimisation. DOI: 10.1504/IJICT.2026.10081306
Abstract: This study evaluates a retrieval-augmented generation (RAG) customer service system on a digital trade platform, addressing bottlenecks in traditional retrieval and generative models. Using a 91-day A/B test (N = 228,059 users), we compared a traditional retrieval-based system with a RAG-enhanced generative system. Results indicate the RAG system significantly improved intent recognition accuracy (from 78.3% to 92.6%) and answer factual accuracy (from 83.1% to 94.5%). First-contact resolution (FCR) increased by 21.7 percentage points, customer satisfaction improved by 0.59 points, and the human transfer rate dropped by 16.4 percentage points. Ablation experiments confirmed the retrieval module minimises hallucinations (2.1%), while the generation module ensures semantic adaptability. This study provides empirical evidence and a three-dimensional evaluation framework encompassing technical performance, user experience, and operational costs for large-scale RAG deployments in digital trade. Keywords: digital trade platform; intelligent customer service; retrieval-augmented generation; RAG; large language models; A/B testing; ablation experiments. DOI: 10.1504/IJICT.2026.10081359
Abstract: Digital platforms are widely used for English learning, yet students often struggle with engagement and progress. This study analyses university students learning behaviours via a learning management system (LMS), using log data on login frequency, quiz scores, and discussion activity. After preprocessing, k-means clustering identified three learner profiles: highly engaged achievers, steady performers, and low-engagement learners. For prediction, an adaptive deep learning framework (EduPre-ADBO-BiGRU) was developed, optimising hyperparameters and achieving over 95.8% accuracy. Results indicate that consistent engagement, frequent review, and multimedia usage correlate with proficiency gains, while irregular participation leads to stagnation. The integration of behavioural analytics, linguistic features, and deep learning offers precise insights for designing adaptive learning pathways and personalised interventions in English education. Keywords: term frequency-inverse document frequency; TF-IDF; education prediction with adaptive dung beetle optimiser-bi-directional gated recurrent unit; EduPre-ADBO-BiGRU; learning management system; LMS; student behaviour. DOI: 10.1504/IJICT.2026.10081381
Abstract: Digital supply chain transformation and environmental, social, and governance performance have complex causal links that traditional correlation analysis cannot clarify. This paper introduces a machine learning-based causal inference and traceability framework integrating structural causal modelling, causal forests, and Shapley values. Using 20182024 panel data from Chinese listed firms, we estimate heterogeneous treatment effects of digital maturity on environmental, social, and governance performance. The framework achieves an area under the curve of 0.892, outperforming traditional models by 9.8% and extreme gradient boosting by 3.4%. Mediation analysis via causal forests shows that digital supplier relationship management mediates 34.7% of the total effect. Shapley-based traceability identifies supplier concentration and digital integration depth as key factors, with significant heterogeneity between state-owned and private enterprises. This study provides a machine learning based causal methodology for sustainable digital supply chains. Keywords: digital supply chain; causal machine learning; causal forest; Shapley value; heterogeneous treatment effect; traceability. DOI: 10.1504/IJICT.2026.10081382
Abstract: This research proposes a conditional generative framework that transforms symbolic music scores into expressive performances using deep learning. The proposed adaptive deer hunting optimised conditional variational auto-encoder (ADHO-CVAE) model incorporates performance context embeddings such as global tempo profiles, expressive style categories, and performer-specific tendencies to produce nuanced variations in timing, dynamics, and articulation not captured in standard notation. Trained on the Maestro-Piano-Midi dataset (N = 1,276) with aligned score-performance pairs, the model learns conditional distributions of expressive features including timing deviation, velocity variation, and articulation ratio. Experimental results show the model outperforms existing approaches, achieving higher fluency (4.6), emotional expression (4.7), and harmony (4.5). The framework effectively bridges symbolic notation and human-like musical interpretation for intelligent music systems. Keywords: expressive music performance; symbolic-to-performance modelling; conditional variational auto-encoders; CVAEs; generative models. DOI: 10.1504/IJICT.2026.10081383
Abstract: Identifying students at risk of academic failure in higher vocational education is essential for timely intervention. Traditional assessment methods often overlook complex behavioural, socio-economic, and institutional factors. This work introduces a data mining-based early warning system (EWS) designed to identify at-risk students through intelligent predictive modelling. The proposed framework combines light gradient CatBoost classifier (LG-CatBoost) for powerful nonlinear risk prediction with artificial protozoa optimiser (APO) for adaptive hyperparameter tuning, resulting in highly accurate identification of at-risk students at an early stage. According to the results, APO-LG-CatBoost model established greater performance when compared to the existing methods, achieving a RMSE of 0.0789 and MAE of 0.0065, indicating high predictive accuracy and stability in estimating student risk levels and it attained a higher accuracy (92%), precision (84%), Recall (86%), and an F1-score (90%), reflecting balanced and reliable at-risk students identification. Keywords: educational data mining; EDM; at-risk students; early warning system; EWS; higher vocational education; predictive analytics. DOI: 10.1504/IJICT.2026.10081384
Abstract: The integration of artificial intelligence (AI) into vocational education has enabled adaptive and data-driven personalisation, fundamentally transforming traditional learning environments. Conventional vocational training systems often fail to accommodate individual learner variability and lack real-time adaptability, leading to uneven skill acquisition, reduced engagement, and limited predictive insight into learner performance. This research introduces an intelligent tutoring system (ITS) for the implementation and evaluation of adaptive vocational education, utilising an adaptive northern goshawk-intelligent residual neural network (ANG-IRNeuroNet) to optimise learning adaptation and performance prediction. The proposed system achieves an overall predictive accuracy above 93%, emphasising its reliability and effectiveness. Overall, the ANG-IRNeuroNet-based ITS establishes a scalable, intelligent, and highly adaptive framework for personalised vocational education, effectively bridging the gap between individualised learning and workforce readiness while ensuring continuous optimisation, learner success, and enhanced educational outcomes. Keywords: artificial intelligence; AI; vocational education; intelligent residual neural network; IRNeuroNet; intelligent tutoring system; ITS. DOI: 10.1504/IJICT.2026.10081385
Abstract: Transmission tower deformation is difficult to monitor continuously because small structural movement is easily hidden by unstable radar scattering and low-coherence observations. To address this problem, this paper proposes a coherence-guided Swin transformer network for deformation monitoring using synthetic aperture radar time-series data. Firstly, tower-centred radar patches are constructed by combining amplitude, phase, coherence and line-of-sight deformation information. Then, hierarchical window attention is used to extract local tower features and neighbouring stable scatterers. Finally, coherence-guided weighting and multi-task prediction are introduced to estimate deformation and identify risk levels. Experimental results show that the proposed method reduces deformation error from 1.05 millimetres to 0.58 millimetres compared with the vision transformer baseline, and improves the risk identification score to 91.8%. The method shows good stability for long corridor tower monitoring under low-coherence conditions. Keywords: transmission tower; deformation monitoring; synthetic aperture radar; Swin transformer; coherence-guided learning. DOI: 10.1504/IJICT.2026.10081386
Abstract: The flow of emotion in musical theatre vocal segments is not a static label, but rather a continuous trajectory that evolves in real time as the drama unfolds. Existing vocal synthesis methods struggle to capture this dynamic rendering, resulting in synthesised voices that lack dramatic tension. This paper proposes a diffusion-based dynamic emotional rendering for musical theatre vocal segments method, which maps script semantics to an arousal-valence emotional time series and uses this to conditionally generate vocal acoustic parameters. Experiments were conducted using two publicly available vocal segment datasets. In emotional consistency evaluation, the proposed method achieved an area under the curve of 0.89, representing a 17% improvement over the baseline model (0.76); normalised discounted cumulative gain @10 improved from 0.72 to 0.84. The results demonstrate that this framework can effectively simulate the dramatic requirements of both gradual and abrupt emotional shifts. Keywords: diffusion models; dynamic emotional rendering; musical theater arias; arousal-valence trajectories; vocal parameter generation. DOI: 10.1504/IJICT.2026.10081387
Abstract: The reinforcement learning approach is used to develop a model that can predict the demand for chronic care services by modelling the disease development process as a Markov decision process where the reward function is subject to delay and decision processes are dependent on each other. Supervised learning techniques fall short of capturing temporal effects and reward structures and hence cannot be efficiently employed to make predictions about chronic care demand in a dynamic setting. To overcome the problem, a policy gradient method for optimisation is employed for training an optimal policy that allocates care resources to patients in different health states. The approach facilitates the adaptive modelling of the evolution of chronic care demand based on the interaction with the environment. The experimental findings show a decrease in prediction error by 18.6% and a gain in healthcare resource allocation efficiency by 14.3% over baseline recurrent neural networks. Keywords: reinforcement learning; chronic care; healthcare prediction; Markov decision progress; policy optimisation. DOI: 10.1504/IJICT.2026.10081388
Abstract: In the automatic scoring of English essays, the existing methods generally have problems of insufficient utilisation of multi-level semantic information and lack of exploration of the relative discriminative ability among samples. To mitigate these problems, this paper suggests a hierarchical semantic network and a contrastive enhancement approach. Firstly, three parallel feature extraction channels of semantics, grammar and text are constructed to achieve multi-dimensional explicit modelling of essay quality. Secondly, an interactive multi-head cross-attention fusion mechanism is designed, which realises the deep integration of heterogeneous features through modal interaction and dynamic weighting. Finally, this paper employs a contrastive learning strategy to boost the models discriminative power across different scoring levels. Experimental outcome indicates that the quadratic weighted kappa of the suggested approach reaches 0.924, which is 0.065 higher than the optimal baseline. This research effectively promotes the development of automatic scoring methods for English essays. Keywords: automatic English essay grading; hierarchical semantic network; contrastive enhancement; multi-head cross-attention; dynamic weighting. DOI: 10.1504/IJICT.2026.10081389
Abstract: In digital cultural heritage preservation, three-dimensional reconstruction often encounters an inherent contradiction between geometric accuracy and visual perception authenticity high-fidelity geometric models lose artistic appearance due to texture distortion, while methods prioritising visual naturalness fail to retain fine structures. This paper proposes weighted hierarchical aggregation, which restores artistic texture continuity while maintaining structural integrity through adaptive weight allocation and hierarchical fusion of multi-scale geometric features. Experiments on standard reconstruction benchmarks show that this method increases structural integrity score to 0.923 (a 3.5% improvement over baseline), texture perception fidelity to 0.894 (a 9.2% gain), and reduces geometric deviation from 1.47 mm to 0.62 mm. This approach fills the gap in existing reconstruction strategies lacking collaborative optimisation between geometric and texture features, providing a viable path for digital art reconstruction that balances quantitative accuracy with subjective visual quality. Keywords: digital art reconstruction; hierarchical feature aggregation; perceptual quality assessment; texture fidelity. DOI: 10.1504/IJICT.2026.10081390
Abstract: Matching degree between digital economy and green policies is hard to quantify because policy vocabulary keeps drifting across planning cycles. To address this problem, this study proposes a temporal word embedding framework for policy matching computation. First, yearly corpus slices are embedded separately and rotated into one shared space through frequency-stable anchors. Then, drift-conditioned term attention aggregates each document into paired semantic and drift representations. Finally, a gated fusion stage weighs the two channels pair by pair and outputs a ranked matching list. Experimental results show that the method reached 0.763 Spearman correlation, 0.804 NDCG, 0.718 precision, and 0.741 MRR on dataset A, with a wider margin on the noisier dataset B. The framework keeps its ranking stable across runs, stays traceable to bridging terms, and reacts to policy convergence about a year before word counts do. Keywords: temporal word embedding; matching degree computation; digital economy policy; green policy. DOI: 10.1504/IJICT.2026.10081391
Abstract: In the social media era, mainstream ideology dissemination faces a key contradiction: information reaches many people, but whether its core meaning remains stable is hard to measure. Existing evaluations rely on behavioral metrics like forwards and likes, answering how many saw it but not how many truly understood it. During dissemination, texts are constantly rewritten, causing original semantics to drift or reverse. Current text similarity methods also lack adaptation to reference framework-specific matching. To address this, this paper proposes the semantic association volatility indicator, transforming effectiveness evaluation into measuring semantic stability over time. Experiments on public corpora show our method achieves 74.2% accuracy in determining effectiveness levels11.4 percentage points higher than static initial-moment methods. The semantic association volatility indicator also correlates at 0.83 with manual effectiveness labels, confirming non-stationary semantic association as a feasible proxy for dissemination effectiveness. Keywords: ideological dissemination; semantic correlation degree; non-stationary time series; communication effectiveness.
Abstract: In the current macro context of global value chain reconfiguration and digital transformation, the intelligent generation and innovation of product shapes are of great significance for enhancing market competitiveness. This paper proposes a product shape design model based on an improved generative adversarial network. This model achieves an effective balance between form innovation and physical entity constraints by introducing specific boundary condition penalty terms and manufacturability indicators in the latent space mapping stage. Experiments show that compared with mainstream baseline methods, the geometric fidelity of this improved model has increased by 5.2%. Moreover, ablation experiments and specific case analyses further confirm that after introducing multi-dimensional constraint feature matching, the incidence of unreasonable topological structures in the model's generated results has decreased by 11.5%. This research provides a unified and practical computational framework for the intelligent design of complex industrial products. Keywords: generative adversarial networks; GANs; product form; latent space; constraint embedding; feature matching.
Abstract: The deployment of multi-modal large models faces a key challenge in heterogeneous computing power environments. To address this, we propose a dynamic adaptation and fine-tuning framework for multi-modal large models (DA-FTMM). In terms of performance, the framework has a BLEU-4 score of 0.35 ± 0. 02, edge latency is reduced by 30% to 280 ms, and energy consumption is reduced by 40%. In terms of robustness, the accuracy decrease rate is only 10 ± 2%, and the recovery time is 5 ± 1 second. The practicality evaluation shows that the user satisfaction is 4.6 ± 0. 2 points and the deployment efficiency improved by 60%. These results verify the advantages of the framework in the balance of accuracy and efficiency. This paper deeply integrates dynamic architecture search and parameter fine-tuning, providing an innovative path for the efficient deployment of multi-modal large models in diverse computing power environments. Keywords: heterogeneous computing power; multi-modal large model; dynamic adaptation; fine-tuning. DOI: 10.1504/IJICT.2026.10081177
Abstract: When virtual power plants integrate dispersed distributed energy sources, they encounter the contradiction of differences in resource response speed and mismatch between the multiple time scales of scheduling instructions. Existing control strategies either fix the resource grouping or separate the day-ahead planning from the intraday correction, resulting in cumulative tracking deviations. This paper proposes partition-hierarchy collaborative, which solves the problem of insufficient coordination between spatial dispersion and time coupling through dynamic partition aggregation and three-level nested optimisation. Simulation results show that the average regulation error of the partition-hierarchy collaborative strategy drops to 0.144 MW, a 16.4% reduction compared to the fixed partition scheme; the defined regulation credibility index reaches 0.87, an increase of 22.5% compared to the scheme without intraday correction. Experiments prove that the collaborative mechanism of dynamic partitioning and rolling correction can simultaneously improve the tracking accuracy and response reliability of the virtual power plant. Keywords: virtual power plant; multi-time scale; synergistic regulation; dynamic zoning. DOI: 10.1504/IJICT.2026.10081140
Abstract: Detecting academic anomalies in higher education is crucial for early intervention in students' academic difficulties. Existing methods struggle to simultaneously capture the dynamic evolution of learning behaviour over time and the complex relationships among various entities, such as students, courses, and instructors. To address this, this paper introduces a framework that integrates an adaptive time-series encoding module with a heterogeneous graph attention network. Experiments on real-world university data demonstrate that the proposed method achieves an area under the curve of 0.913, representing a 6.2% improvement over the best baseline; its precision is 0.847 and recall is 0.835, both outperforming the comparison models. The method does not require predefined fixed time windows or simplified relationship types, and can directly learn discriminative features from raw trajectories and heterogeneous networks. Keywords: academic performance monitoring; adaptive time-series modelling; heterogeneous graph attention. DOI: 10.1504/IJICT.2026.10081142
Abstract: In online learning, learners' behavioural sequences are highly dynamic. Existing prediction models often treat behaviours at different time points as independent features, ignoring temporal dependencies, which leads to delayed risk identification and poor interpretability. This paper proposes the interpretable prediction and intervention framework for online learning. By learning longterm dependency patterns in behavioural trajectories, the framework provides early warning three weeks before a risk occurs. Experiments show that the method raises the risk identification recall rate to 89.1%, which is 7.2 percentage points higher than the baseline bidirectional long shortterm memory model. The opening rate of intervention push messages reaches 73.8%, significantly above the typical level in similar studies (less than 50%). This research fills the gap of interpretable temporal attribution in online learning warning, enabling teachers to trace back to which week and which behaviour pattern triggered the alert, and thus formulate targeted intervention strategies. Keywords: online learning; time series prediction; explainable artificial intelligence; educational intervention. DOI: 10.1504/IJICT.2026.10081143
Abstract: Government procurement continues to grow in scale, while bid-rigging grows more covert and organised - posing a persistent challenge for audit supervision. Existing methods rely largely on manual rule screening or single-source mining, which fail to reveal implicit links among bidders across data sources, leading to high miss rates and weak interpretability. This paper proposes a bid-rigging risk identification mechanism built on multi-source knowledge graphs. We first integrate heterogeneous audit data through a cross-source entity alignment method to construct a procurement audit knowledge graph. We then design an association-aware heterogeneous graph representation model that aggregates structural links via relation-level attention, jointly encodes quotation and temporal behaviour features, and fuses them through a gating mechanism. Finally, a gang-level scoring and joint optimisation algorithm enables end-to-end inference from individual links to collusion risk. Experiments show our method attains an F1 of 0.913, exceeding the best baseline by 6.2 points. Keywords: government procurement audit; knowledge graph; bid-rigging; graph representation learning; risk identification. DOI: 10.1504/IJICT.2026.10081144 |
Open Access
Keep up-to-date
-
Our Blog -
Join us on Bluesky -
Follow us on X -
Visit us on Facebook - Our Newsletter (subscribe for free)
- RSS Feeds
- New issue alerts
- Inderscience is a member of publishing organisations including:
-
-
