International Journal of Information and Communication Technology (16 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
   
  
• 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
   
  

• Open Access

  Personalising physical education courses via predictive analytics to improve student engagement and health outcomes ( Free Full-text Access )

  by Can Li 
  Abstract: Traditional physical education (PE) often adopts a one-size-fits-all approach, limiting student engagement (SE) and health outcomes. This study proposes a dynamic roach infestation-driven weighted random forest (DRI-WeightedRF) model that integrates swarm intelligence into ensemble learning to predict SE and health outcomes. In DRI-WeightedRF, cockroach-inspired exploration behaviours dynamically optimise feature weights and decision-tree importance: engagement-seeking, wellness motivation, vigilance, and collaborative peer learning guide adaptive weighting to capture complex physiological-psychological interactions. Unlike conventional optimisation-based Random Forest methods that statically tune hyperparameters, DRI-WeightedRF continuously refines feature relevance and tree contributions, enhancing predictive accuracy. Evaluated on a dataset of 5,000 collegiate athletes, DRI-WeightedRF achieved an accuracy of 98.50% with RMSE 0.0863, MSE 0.0074, and MAE 0.0421, outperforming existing models. The results demonstrate the models efficacy in personalising PE interventions to improve engagement and wellness.
  Keywords: physical education; student engagement; health outcomes; dynamic roach infestation-driven weighted random forest; DRI-WeightedRF.
  DOI: 10.1504/IJICT.2026.10081602
   
  

• Open Access

  A deep reinforcement learning model for dynamic generation of ideological and political teaching content ( Free Full-text Access )

  by Xiangqin Zhou 
  Abstract: To address the misalignment of static pre-designed teaching content with learners dynamic needs in ideological and political education, this paper proposes a dynamic content generation model based on deep reinforcement learning. The model formulates the sequential decision-making process of generating teaching content as a Markov decision process and designs a dual-encoder deep q-network algorithm. By integrating a gating mechanism with dual encoders for content semantics and pedagogical suitability, along with knowledge graph constraints, the model achieves real-time optimisation and value-oriented control of content sequences. Experiments on public datasets show that compared to standard deep q-network and collaborative filtering baselines, the proposed model improves content matching accuracy to 76.3% (an 8.0-percentage-point improvement over the standard deep Q-network) and cumulative training reward to 78.5 (a 10.3% relative improvement over proximal policy optimisation), validating its dynamic adaptation strategy.
  Keywords: deep reinforcement learning; ideological and political education; teaching content generation; deep Q-network; DQN.
  DOI: 10.1504/IJICT.2026.10081619
   
  

• Open Access

  Dynamic perception of ideological and political identity in classroom settings via multimodal self-supervised representation learning ( Free Full-text Access )

  by Fei Wang 
  Abstract: Ideological and political identity is a dynamic process continuously constructed in classroom teaching, whose perception faces three challenges: heterogeneous modalities are difficult to align, annotated samples are scarce, and signals fluctuate. This paper proposes a dynamic perception method for ideological and political identity based on multimodal self-supervised representation. The method designs two pre-training tasks: temporal consistency prediction and mask reconstruction, to learn high-discriminative representations under unlabeled conditions; it achieves unified heterogeneous modality space through progressive semantic alignment, introduces confidence gate adaptive adjustment to fusion weights to cope with signal fluctuations, and adopts gated recurrent units to model the temporal accumulation effect of identity, outputting the three-dimensional intensity of cognition, emotion, and behaviour. Experimental results show that the proposed method achieves an accuracy of 92.1% and a macro average F1 of 91.9%, which is significantly better than four baseline methods, verifying its effectiveness in classroom emotion perception tasks.
  Keywords: ideological and political identity perception; multimodal fusion; self-supervised representation; semantic alignment; gated recurrent unit.
  DOI: 10.1504/IJICT.2026.10081620
   
  

• Open Access

  A semantic embedding framework for analysing discourse patterns in English teaching ( Free Full-text Access )

  by Yibin Huan 
  Abstract: Teaching discourse serves as a core classroom teaching carrier, whose structure and semantics impact teaching organisation and learning comprehension. Existing methods, relying on manual rules and qualitative annotations, struggle to model large-scale, multi-style teaching corpora. This paper proposes a method integrating BERT-based context-aware semantic embedding, sliding window-enhanced semantic continuity modelling, semi-supervised functional segment clustering, and a graph mechanism combining semantic similarity and time sequence to restore teaching logic evolution. Tested on three diverse English classroom datasets, the method outperforms baselines in clustering accuracy, style adaptability, and structural clarity, verifying its effectiveness and generalisability. Results support intelligent teaching tasks like discourse comprehension and personalised feedback generation.
  Keywords: English teaching discourse; semantic embedding; BERT; structural modelling; graph analysis.
  DOI: 10.1504/IJICT.2026.10081695
   
  

• Open Access

  Personalised recommendations for sports tourism resources based on collaborative filtering algorithms ( Free Full-text Access )

  by Zongji Zhao 
  Abstract: The recommendation of sports tourism resources has long faced a core dilemma: the system knows what users like, but cannot determine what intensity they can handle. This mismatch leads to recommendations that either exceed the users physical limit or feel boring. This paper proposes physical activity and seasonality integrated collaborative filtering. It embeds user movement ability and seasonal preferences into the traditional rating set, enabling the system to distinguish beginners from experts. Experiments show that physical activity and seasonality integrated collaborative filtering improves recommendation accuracy from 0.150 to 0.182 and intensity matching degree from 0.626 to 0.742. Essentially, this work fills the ability perception gap in sports tourism recommendation allowing the algorithm to understand the users desire while respecting their physical boundaries.
  Keywords: collaborative filtering; sports tourism; intensity adaptation; seasonal constraints.
  DOI: 10.1504/IJICT.2026.10081696
   
  

• Open Access

  Geological hazard monitoring and risk early warning of transmission corridor based on multi-modal remote sensing data ( Free Full-text Access )

  by Siqing Xu, Yulu Ren, Hui Yuan 
  Abstract: In the monitoring of geological disasters in power transmission corridors, multi-modal remote sensing data often suffer from significant differences in imaging mechanisms, being frequently obscured by clouds and fog, as well as being disturbed by coherent speckle noise and other background interference, resulting in low accuracy of disaster monitoring and risk warning. To tackle this challenge, the proposed method first extracts complementary characteristics from each modality via a multi-branch feature extraction module, and then achieves adaptive cross-modal alignment through a context-guided spatial characteristic fusion module. Next, the rectangular self-calibration module dynamically broadens the receptive field, enabling robust alignment with high-aspect-ratio, strip-like objects. Additionally, the joint loss function is combined to achieve end-to-end collaborative optimisation for disaster monitoring and risk warning. Experimental findings demonstrate that the suggested approach achieves an average accuracy improvement of at least 3.69% over the baseline approach, confirming its effectiveness.
  Keywords: perception-guided; diffusion model; colour combination generation; colour perception alignment; fashion design.
  DOI: 10.1504/IJICT.2026.10081771
   
  

• Open Access

  Deep Q-network-based intelligent financial decision-making model for enterprise dynamic budget allocation ( Free Full-text Access )

  by Sijia Zhang 
  Abstract: The deep Q-network (DQN) methods usually suffer from the convergence failure problem for enterprise dynamic budget allocation due to the credit allocation ambiguity and the multi-objective constraints. Long-horizon feedback can cause myopic policies or the model to be noise-sensitive. To solve this problem, a DQN model based on potential energy shaping and composite penalty terms is proposed. A potential energy function based on a health index captures sparse profit signals into dense process feedback, effectively mitigating delayed credit assignment. Moreover, the risk- and volatility-based penalty terms encourage the model to find stable and reliable decision-making solutions by discouraging high-variance strategies. The experimental results show that when the penalty coefficient is set to 1.5, the proposed model can achieve risk-adjusted returns of 0.93 and 1.92 in low-noise conditions and the return decreases by only 31.3% in high-noise conditions, indicating strong robustness and adaptability.
  Keywords: deep Q-network; DQN; dynamic budget allocation; potential-based reward shaping; multi-objective constraint; risk-adjusted return.
  DOI: 10.1504/IJICT.2026.10081772
   
  

• Open Access

  Multi-modal alignment-driven consistency verification and error traceability for financial statement accounts ( Free Full-text Access )

  by Jingfan Zhao 
  Abstract: Financial statements combine text annotations, table structures, and numerical values, making automated verification difficult. A unified framework is presented that maps these three modalities into a shared embedding space through contrastive learning, addressing a gap left by prior multimodal models that align modalities only at the output layer and offer no built-in traceability. A consistency network predicts the verification score, and an attention-based tracing module localises the source modality and account of any detected exception. On a curated dataset of 12,450 statements from 380 listed companies, the proposed framework attains precision of 95.2%, recall of 93.8%, F-measure of 94.5%, exceeding the strongest baseline by 6.5 percentage points. The tracing module further reduces manual investigation time by an average of 61.4%, confirming that joint cross-modal alignment substantially improves both detection and traceability of financial-statement inconsistencies.
  Keywords: multimodal alignment; financial statements; error tracing; consistency verification; cross-modal learning.
  DOI: 10.1504/IJICT.2026.10081773
   
  

• Open Access

  Intelligent sequencing of intangible cultural heritage labour course content via knowledge graph embedding ( Free Full-text Access )

  by Qiang Liu, Qide Lin, Qinghuan Yu 
  Abstract: The content of intangible cultural heritage skills training courses relies on experience-based sequencing, facing challenges such as fragmented knowledge and a lack of personalisation. This paper proposes the knowledge graph-enhanced reinforcement learning content sequencing algorithm, which constructs a four-layer semantic graph encompassing skills, tools, processes, and culture, thereby transforming content sequencing into sequential decision-making. On an intangible cultural heritage dataset, this method achieved an accuracy rate 6.2% higher than the best baseline, with an area under the curve of 0.917; precision and recall increased by 5.8% and 4.8%, respectively, while normalised discounted cumulative gain @10 improved by 7.39%. The results demonstrate that the structural prior of the knowledge graph, combined with the dynamic adaptation of reinforcement learning, can effectively overcome the limitations of traditional methods which overlook the intrinsic connections among skills and the cognitive state of learners and provide a data-driven, computable solution for organising intangible cultural heritage curriculum content.
  Keywords: knowledge graph; reinforcement learning; RL; intangible cultural heritage labour courses; content curation; personalised learning.
  DOI: 10.1504/IJICT.2026.10081774
   
  

• Open Access

  Interactive violin music generation using LSTM and transformer ( Free Full-text Access )

  by Jian Song, Chen Liang 
  Abstract: Violin music generation faces the core challenges of long-range structure and local expressiveness. Pure long short-term memorys struggle to maintain clear harmonic coherence when modelling dependencies in musical sequences, while pure Transformers, though capable of capturing the global framework, produce stiff results in details such as vibrato and glissando. This paper proposes a hybrid generative architecture that combines a transformer encoder with a long short-term memory decoder and incorporates interactive feedback on user preferences, enabling the generation process to be fine-tuned according to the context. On a labelled dataset comprising classical violin works, the hybrid model improved note accuracy from 83.2% to 91.7% compared to a pure long short-term memory, and increased recall from 79.8% to 88.4% compared to a pure long short-term memory. The area under the curve metric also showed improvement, demonstrating the synergy between the hybrid architecture and the interactive strategy.
  Keywords: violin music generation; hybrid neural network architecture; interactive feedback optimisation; symbolic sequence modelling.
  DOI: 10.1504/IJICT.2026.10081775
   
  

• Open Access

  Large language model driven construction and intervention of conceptual metaphor network in second language vocabulary ( Free Full-text Access )

  by Gang Shen, Tao Feng, Hanyuan Zhao 
  Abstract: Abstract vocabulary acquisition in second language suffers from the lack of metaphorical competence. Existing systems only classify and recognise abstract words, ignoring mapping relationships and cognitive load. This study proposes a metaphor network construction method for cognitive load optimisation, which weaves a concept map through the graph attention network, and dynamically adapts the scaffold for learners. Experiments show that the model improves by 4.7 and 3.4 percentage points compared with the optimal baseline. After six weeks of intervention, the metaphor density of learners was increased by 125%, the vocabulary retention rate was 78.4%, 18.2 percentage points higher than that of the control group, and the cognitive load was significantly decreased. The results verify the effectiveness of the path in second language teaching and provide a new idea for the intelligent tutoring system; ITS.
  Keywords: conceptual metaphor network; large language model; LLM; graph attention network; GAN; second language vocabulary acquisition.
  DOI: 10.1504/IJICT.2026.10081776
   
  

• Open Access

  Construction and verification of real-time interactive facial expression synthesis technology for three-dimensional animation driven by fuzzy logic interpolation algorithm ( Free Full-text Access )

  by Lu Zhang, Yachao Jie, Junyao Wang 
  Abstract: This study formalises real-time expression synthesis as a constrained dynamic interpolation problem by modelling the 3D facial expression parameter space. A fuzzy logic inference mechanism is introduced to realise the adaptive adjustment of interpolation weight and smoothing strength. The experimental results show that the expression smoothness index can be as low as 0.167, the time stability score can be as high as 0.884, and the high-frequency jitter suppression ratio exceeds 69% under various interactive data subsets, all of which are better than the comparison method. The ablation experiment further verified the key role of the fuzzy logic inference module and adaptive smoothing mechanism in improving the stability of expression. To sum up, the proposed fuzzy logic interpolation-based facial expression synthesis performs well in real-time, stability, and engineering, providing a valuable reference for 3D animation real-time interaction and expression synthesis.
  Keywords: three-dimensional facial expression synthesis; fuzzy logic; real-time interaction; expression interpolation; stability control.
  
  

• Open Access

  A graph neural network-based algorithm for planning and recommending learning pathways in civil engineering education ( Free Full-text Access )

  by Tengfei Liang, Ruijuan Yang, Xine Yan 
  Abstract: Civil engineering education demands sequenced learning pathways due to complex prerequisite dependencies spanning six civil engineering sub-disciplines. Existing recommendation approaches treat curricula as flat sequences, failing to capture multi-directional knowledge dependencies. This paper proposes a learning pathway graph network that represents curriculum structure as a knowledge graph and models student progression through a student state graph. A dual graph neural network architecture comprising a curriculum encoder and a student state encoder encodes curriculum dependencies and individual learning states simultaneously, and a policy gradient reinforcement learning algorithm optimises recommendations through a composite reward function to maximise long-term outcomes. Evaluated on 3,840 civil engineering undergraduates, the framework achieves improvements of 8.3 percentage points in pathway coherence, 11.7 percentage points in completion rate, and 7.2 percentage points in knowledge mastery over the strongest baseline, confirming that jointly encoding curriculum topology with student state substantially enhances personalised recommendation.
  Keywords: learning path; graph neural network; GNN; civil engineering education; knowledge map; recommended algorithm.