International Journal of Information Quality (13 papers in press)

Regular Issues

• Spatio-Temporal Attention LSTM Model for Tourist Behaviour Prediction and Smart Tourism Route Optimisation  
  by Hongling Wang 
  Abstract: To address insufficient tourist behaviour prediction accuracy and poor path planning in smart tourism, this paper proposes a CNN-LSTM-Attention joint model integrating CNN, LSTM, and attention mechanisms. Traditional methods struggle with spatiotemporal nonlinear features, dynamic congestion, and multi-objective optimisation- failing to accurately predict peak-hour passenger flow or generate personalised routes. This study uses an ST-CNN to extract local/global scenic spot transition features, combines CBAM for dynamic spatiotemporal weighting, designs a multi-task LSTM framework integrating personalised, environmental, and real-time congestion data, and develops a Pareto improvement-based path optimisation algorithm. Experimental results show the model outperforms existing methods in four indicators (ST-RMSE: 8.3, DTW: 9.3, PHR: 0.88, POG: 1.38), reducing spatiotemporal prediction error by 22.4% and improving path optimisation gain by 10.4%, offering an integrated solution for smart tourisms high-precision prediction and interpretable decision-making.
  Keywords: Convolutional neural networks; long short-term memory networks; attention mechanisms; spatiotemporal convolutional network.
  DOI: 10.1504/IJIQ.2026.10077816
   
  
• Optimisation Method for Intelligent Scheduling System of Coal Mine Auxiliary Transportation via Multi-Objective Reinforcement Learning  
  by Tao Lian, Haidong Hu, Can Guo, Yubao Guo, Nan Xu 
  Abstract: To address the low efficiency and insufficient practical constraint consideration in coal mine intelligent trackless auxiliary transportation scheduling, this study proposes a multi-objective twin delayed deep deterministic policy gradient algorithm (MOTD3). Integrating deep reinforcement learning, the algorithm optimizes three core objectives: minimizing transportation costs, electric locomotive robot waiting time, and transportation expected deviations, while accounting for constraints like loading/unloading capacities and time limits. MOTD3 adopts twin critic networks, target policy smoothing, and adaptive experience replay to enhance stability and exploration efficiency. Experimental results show it outperforms traditional evolutionary and reinforcement learning baselines, reducing transportation costs by 42.9% and improving locomotive utilization by 33.5%, offering impactful practical value for intelligent coal mine systems.
  Keywords: Deep reinforcement learning; Multi objective optimization; Intelligent trackless auxiliary transportation; Scheduling optimization.
  DOI: 10.1504/IJIQ.2026.10078394
   
  
• A Virtual Power Plant Electricity-Carbon Price Forecasting Model Based on Adaptive Learning Algorithms  
  by Gang Ma, Zhaohua Zhang, Yan Zhang, Juncheng Guang, Yiqun Zhou, Yawei Mu 
  Abstract: As a core platform for distributed energy integration, virtual power plants (VPPs) face dual uncertainties from deep electricity-carbon market coupling. To optimize energy allocation and mitigate low-carbon transition risks, this study proposes an electricity-carbon price forecasting framework based on adaptive learning algorithms. It adopts an exponential dynamic weighting strategy, integrates an enhanced zebra optimization algorithm with golden sine search and KNN feature selection to identify key price drivers, and uses a genetic algorithm to fine-tune BP neural network weight matrices and bias vectors for fitting complex nonlinear price patterns better. Experiments on GEFCom 2014 show MAE, MAPE, RMSE improved by 0.82%, 2.75%, 2.41% vs. benchmarks (XGBoost, BP-SAA, BP-PSOA). On EPE, it achieves MAE 0.12%, MAPE 0.52%, RMSE 1.13%, outperforming all comparators, verifying its high reliability in complex environments and supporting VPPs’ low-carbon risk mitigation and efficient operation.
  Keywords: Electricity-carbon price forecasting; BP neural network; Zebra optimization algorithm; Genetic algorithm; KNN feature selection.
  DOI: 10.1504/IJIQ.2026.10078395
   
  
• Differentiable Neural Architecture Search Variational Autoencoder Model for Vocal Style Transfer  
  by Jingjing Wang, Muhammad Attique Khan 
  Abstract: This study proposes the MUDAS-Net framework, which integrates Differentiable Neural Architecture Search (DARTS), Adaptive Instance Normalization (AdaIN), and a style embedding mechanism for efficient vocal style transfer. Firstly, MUDAS-Net leverages DARTS for automatic network architecture optimization, generating the optimal model architecture tailored for the vocal style transfer task, thereby enhancing the transfer performance. Then, the AdaIN technique is applied to finely adjust content and style information, enabling precise injection of the target style while preserving the melody and lyrics of the source audio. Finally, the style embedding module further enhances the model's adaptability to different singers and styles. Experimental results demonstrate that MUDAS-Net significantly outperforms traditional vocal style transfer methods on multiple benchmark datasets and real-world singing data, achieving excellent results in key metrics such as audio quality, style consistency, and naturalness. This study offers strong technical support for future personalized speech synthesis and voice style conversion technologies.
  Keywords: Singing Voice Style Transfer; DARTS; AdaIN; Neural Architecture Search.
  DOI: 10.1504/IJIQ.2026.10078715
   
  
• Sports Motion Trajectory Extraction Network Based on TCN-STA-GRU  
  by Jialin Li, Alireza Sharifi, Mohammadmahdi Safari 
  Abstract: Sports motion trajectory extraction is crucial for training guidance and tactical analysis, yet existing methods often struggle to capture discriminative trajectory patterns and model temporal dependencies in complex scenarios. This paper proposes a TCN-STA-GRU network for trajectory extraction and prediction. First, a multi-scale Temporal Convolutional Network (TCN) with dilated causal convolutions and residual connections captures motion cues across different temporal ranges. Second, a Spatio-Temporal Attention (STA) module highlights salient information in both space and time: the spatial branch localizes key positions, while the temporal branch emphasizes critical moments, suppressing background noise and irrelevant details. Finally, a Gated Recurrent Unit (GRU) performs deep temporal modeling via update and reset gates to learn long-term dependencies in trajectories. Experiments on three sports datasets (Basketball-D, Football-D, and Soccer-D) demonstrate superior performance. On Basketball-D, the proposed method achieves a 96.0% F1-score and a 1.89 average displacement error, confirming its effectiveness.
  Keywords: Motion trajectory extraction; Temporal Convolutional Network; Gated Recurrent Unit; Spatio-Temporal Attention Mechanism; Temporal feature modeling; Deep learning.
  DOI: 10.1504/IJIQ.2026.10078725
   
  
• Smart Contract-Driven Pricing Strategy for Web3 Crowdfunding Based on Multimodal Deep Clustering  
  by Xiang Chen, Kan Lu, Alpamis Kutlimuratov 
  Abstract: In response to the challenges posed by dynamic market fluctuations and contributor heterogeneity in Web3-based crowdfunding platforms, this study proposes a novel pricing strategy underpinned by multimodal deep clustering and smart contract automation. By integrating behavioral, financial, and social data from both on-chain and off-chain sources, the proposed framework employs autoencoder-based deep clustering and contrastive learning to segment contributors into latent pricing clusters. These segments inform a dynamic pricing policy executed through EVM-compatible smart contracts. Extensive experiments on simulated and benchmark datasets demonstrate superior clustering performance and pricing effectiveness, including increased funding completion rate (+13.3%), early-stage participation (+21.3%), and reduced price volatility. System-level evaluations further validate the model's feasibility in decentralized settings, maintaining sub-second latency and >94% contract execution success under high load. This research advances the theoretical and practical foundations of adaptive pricing strategies in decentralized economies by fusing AI-driven segmentation with blockchain-native automation.
  Keywords: Web3 crowdfunding; deep clustering; multimodal fusion; contrastive learning; dynamic pricing; smart contract.
  DOI: 10.1504/IJIQ.2026.10078728
   
  
• Community-Guided Cascade Fusion in Information Diffusion Prediction  
  by Yingting Lin, Fei Xiong, Zhiyuan Zhang 
  Abstract: The objective of information diffusion prediction is to estimate the probability that inactive users in a cascade will be activated. Existing studies either capture the sequence dependencies within the cascade or leverage user graph structures to predict future activations. However, existing methods overlook users’ positional roles and community structures, and typically rely on a single cascade structure, limiting their ability to jointly model sequential and non-sequential patterns. To address these limitations, we propose a Community-Guided Cascade Fusion Network (CGCF-Net). Specifically, we employ the Louvain algorithm to identify potential community relationships among users and compute the speed and position embeddings of users within each community to obtain their community representations. Furthermore, we leverage coverage attention mechanism and GRU to capture different structural representations, which are fused to obtain the final cascade representation. Extensive experiments on three public datasets verify the effectiveness of the proposed model.
  Keywords: information diffusion; social network analysis; attention; neural network; hypergraph; community analysis; data mining.
  DOI: 10.1504/IJIQ.2026.10078795
   
  
• Multimodal Data Mining in Learning Processes: Optimising the Allocation of Online English Teaching Resources  
  by Le Yang 
  Abstract: In English online teaching, the insufficient utilisation of multimodal data hinders the effectiveness of resource optimisation and allocation. Therefore, a research on optimising the allocation of English online teaching resources under multimodal data mining in the learning process is proposed. Principal component analysis and local outlier factor algorithm are used to process multimodal data, and feature mining and fusion are carried out through self-attention mechanism and graph neural network. Based on the feature fusion results, utilise long short-term memory networks to achieve resource demand prediction. Finally, the objective function for optimising the allocation of English online teaching resources is constructed, and the bat algorithm is used to solve it and obtain the optimal optimisation configuration scheme. The test results show that the proposed methods have a time consumption of less than 3.20 s, a maximum configuration accuracy of 0.97, and a skill mastery rate of over 83% for students.
  Keywords: English online teaching; Learning process; Multimodal data mining; optimized allocation of resources.
  DOI: 10.1504/IJIQ.2026.10078799
   
  
• Terminal fault processing in power system based on reinforcement learning and mobile edge computing  
  by Jing Yang, Qiang Song, Qingqing Fu 
  Abstract: To address the critical need for efficient terminal fault handling in power systems, this study proposes an integrated framework that combines reinforcement learning with Mobile Edge Computing (MEC). A Markov decision model is constructed to formalize fault processing as a sequential decision problem, and an enhanced Deep Deterministic Policy Gradient algorithm with stochastic variance reduction gradient (DDPG-SVR) is developed to mitigate gradient estimation biases in traditional DDPG approaches. Experimental evaluations show that the proposed method yields close prediction fit for key transformer parameters, and achieves detection precision of 91.5% and 90.4% for complete failure and deviation faults, respectively. Meanwhile, the MEC-enabled design effectively reduces fault processing latency and edge node energy consumption. These results validate that the proposed framework significantly improves the efficiency, real-time performance and energy economy of terminal fault processing, supporting stable and reliable operation of power systems.
  Keywords: reinforcement learning; mobile edge computing; Markov decision making; stochastic variance reduction.
  DOI: 10.1504/IJIQ.2026.10078910
   
  
• Integration of Multimodal Ideological and Political Teaching Materials Based on Improved Fuzzy Clustering Algorithm  
  by Jianwei Zhao 
  Abstract: Studying the integration method of multimodal ideological and political teaching materials is of great significance for promoting the deep integration of information technology and ideological and political education. Therefore, a new integration method of multimodal ideological and political teaching materials based on improved fuzzy clustering algorithm is proposed. Constructing a multimodal feature extraction and fusion model, utilizing collaborative attention mechanism to achieve deep fusion of image and text features; Introducing an improved bat algorithm to optimize the initial cluster center selection of fuzzy clustering algorithm, and reconstructing the objective function that integrates sample and attribute weights, to improve the accuracy and stability of multimodal ideological and political teaching material integration. The experimental results show that the proposed method has good multimodal feature extraction effect on ideological and political teaching materials, good inter class separation, and almost no obvious outliers or misclassified points, the highest ARI reached 0.96
  Keywords: Multimodal; Ideological and political education; Fuzzy clustering; Bat algorithm; Feature; Objective function.
  DOI: 10.1504/IJIQ.2026.10078914
   
  
• Pyramid-aware feature fluctuation U-Net for medical image segmentation and content-based retrieval  
  by Chunfa Wu, Dongdong Chen, Hongqing Zheng, Yangpeng Huang, Muhammad Haris 
  Abstract: This study proposes Feature Fluctuation U-Net (FFU-Net), a novel deep learning framework designed to improve the quality of medical image analysis for segmentation and content-based retrieval in remote healthcare education. FFU-Net adopts ResNet34 as the encoder with atrous convolution for fine-grained feature extraction, embeds three Advanced Pyramid Transduction modules to enhance global contextual information, and integrates a Feature Fluctuation Pyramid Module for multi-scale feature interaction. Experiments on public LiTS, 3Dircadb, and CHAOS datasets demonstrate it outperforms benchmark models in key segmentation metrics and achieves 96.2% accuracy in abdominal CT retrieval. This framework effectively supports remote medical education by bridging clinical imaging resources and educational information demands.
  Keywords: medical image segmentation; content-based retrieval; U-Net; feature fluctuation; remote healthcare education; deep learning.
  DOI: 10.1504/IJIQ.2026.10078962
   
  
• Study on Data-Driven Approach to Optimizing Interactive Online English Teaching Across Multiple Platforms  
  by Wenxin Wang 
  Abstract: To tackle data fragmentation in multi-platform English online teaching, this study proposes an optimization method using collaborative data mining. A distributed data collection system based on the Scrapy framework integrates multi-source data from LMS, interactive, social, and learning platforms. The CEEMDAN algorithm denoises data, while LSTM captures temporal features of student behaviors. A PSO-BestK hybrid clustering algorithm enables precise student grouping. Based on clustering results, data-driven optimization strategies are introduced, including differentiated interaction design, dynamic teaching adjustments, and innovative teacher-student interaction modes. Experiments show the method achieves over 90% accuracy in student recognition, improves learning performance by more than 16%, and significantly increases classroom interaction frequency
  Keywords: Multi platform collaborative data; Data mining; English online interactive teaching; Optimization of teaching mode; PSO BestK hybrid clustering algorithm.
  DOI: 10.1504/IJIQ.2026.10079014
   
  
• An Adaptive Transformer-Based Framework for Music Classroom Interaction Assessment  
  by Zhaoyu Zhu, Shuang Wu, Wahab Khan 
  Abstract: Music classroom interaction assessment is critical for intelligent education and educational quality development, yet existing methods suffer from flaws in multimodal data fusion, long-term temporal dependence modeling and multi-dimensional evaluation. This paper proposes an adaptive Transformer-based assessment method with three innovations: a Dynamic Adaptive Attention Fusion module (DAAF) that adaptively allocates attention weights across audio, video and text modalities for collaborative multimodal feature modeling; a Hierarchical Temporal Transformer (HTT) with a local-segment-global structure to capture beat-, segment- and classroom-level interaction features; and a Multi-Task Joint Evaluation Framework (MTJEF) with shared encoders and task-specific decoders for joint evaluation of student involvement, interaction frequency, emotional state and music skill feedback. Experiments on MUSIC-AVQA and DAiSEE datasets show the method boosts accuracy by 7.9% and reduces MAE by 22.1% compared with baselines.
  Keywords: transformer; music classroom; interaction assessment; multimodal fusion; multi-task learning; attention mechanism.
  DOI: 10.1504/IJIQ.2026.10079179