Forthcoming Articles
International Journal of Data Mining and Bioinformatics
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 Articles will not be published until the final proofs are validated by their authors.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 Data Mining and Bioinformatics (26 papers in press) Regular Issues
Abstract: Endometrial cancer exhibits substantial prognostic heterogeneity, highlighting the need for accurate survival prediction. This study developed a deep learning-based multimodal fusion model integrating clinical, imaging, pathological, and molecular data. Convolutional neural networks extracted imaging features, while multilayer perceptrons encoded clinical and molecular information. An attention mechanism enabled weighted feature fusion, followed by a DeepSurv model for individualised survival prediction. Model performance was evaluated against Cox proportional hazards and random survival forest (RSF) models. The proposed model achieved a C-index of 0.802 (95% CI: 0.7560.848), outperforming the Cox (0.682) and RSF (0.716) models. The 12-month area under the curve reached 0.802, while the 60-month Brier score was 0.142. These findings demonstrate the potential of multimodal deep learning to improve survival prediction and support risk stratification and personalised treatment decisions in patients with endometrial cancer. Keywords: deep learning algorithm; multimodal medical data; feature fusion; DeepSurv model; endometrial cancer prognosis. DOI: 10.1504/IJDMB.2026.10081815 Special Issue on: Hyperautomation and Big Data Methodologies Interactivity and Applications
Abstract: The social media platforms used by museums face various challenges, such as inaccurate generation of cultural information, difficulty in accurately matching changes in user interests, and insufficient ability to adapt to immersive cross platform experiences. To address these issues, this study proposes an intelligent content generation and personalised push algorithm framework. This paper first develops an intelligent content generation model based on graph convolutional network (GCN) transformer, then proposes a dynamic interest model that includes context aware users, and finally proposes a push decision mechanism based on multi-objective particle swarm optimisation deep Q network (MOPSO-DQN). The experiment shows that the average historical matching accuracy of the GCN Transformer model is 92.27%, the cultural accuracy is 95.35%, the semantic consistency is 94.38%, 2 S. Wang et al. and the factual accuracy is 96.79%. The algorithm framework proposed in this study provides a reliable technical solution for the intelligent and digital distribution of museums. Keywords: social networking technology; museum social media; knowledge graph; KG; content generation; personalised recommendation; context-aware interest modelling. DOI: 10.1504/IJDMB.2026.10078865
Abstract: Enterprise decision-making increasingly depends on data that are high-dimensional, time-varying, and strongly connected across departments, agents, and operational processes. Conventional analytics models often treat these data as static or independent records, which limits their ability to capture changing relationships and support timely decisions. To address this limitation, we propose a deep learning-based graph framework for enterprise decision support. The proposed dynamic enterprise synthesis graph (DESG) represents enterprise agents, resources, and operational states as nodes and links in a dynamic graph, allowing the model to learn both temporal changes and relational dependencies. On top of this graph representation, we develop an adaptive multi-agent intent orchestration (AMAIO) module that adjusts agent-level decisions according to local states, learned interaction weights, and global enterprise objectives. In this way, the framework connects local operational behaviour with overall decision goals while remaining adaptable to partially observable and resource-constrained environments. Experiments on multiple benchmark datasets show that the proposed method achieves better accuracy, recall, F1-score, and AUC than representative recurrent, convolutional, and transformer-based baselines. The results indicate that combining dynamic graph modelling with adaptive multi-agent coordination can improve the robustness, coherence, and efficiency of enterprise-scale decision-making systems. Keywords: multi-agent systems; temporal graph models; multi-agent decision coordination; computational enterprise analytics; intelligent decision support. DOI: 10.1504/IJDMB.2026.10081098
Abstract: Enterprise management decision making requires the integration of diverse data modalities, including financial metrics, operational statistics, and external market indicators. The heterogeneity and dynamic characteristics of these modalities create substantial challenges for effective decision support. Traditional approaches often fail to capture complex interactions among multiple modalities, resulting in limited predictive accuracy and insufficient interpretability. To address these limitations, this study introduces M3DecideNet, a multi modal attention driven fusion framework for enterprise management decision support. M3DecideNet serves as the overall decision support framework and is organised through three hierarchical components. A formalised problem definition establishes the mathematical foundation for multi modal decision modelling. M3FusionNet functions as the core modelling component within M3DecideNet, extracting modality specific features and integrating them through attention-based weighting to generate a unified representation. The adaptive decision fusion strategy, namely ADFS, operates as the decision refinement layer by introducing context aware modulation and regularisation to improve robustness, adaptability, and interpretability. Empirical evaluations show that M3DecideNet achieves superior predictive performance compared with existing methods while providing scalable and actionable insights across diverse enterprise scenarios. The proposed framework offers a structured, robust, and interpretable solution for modern enterprise management decision making. Keywords: enterprise management decision; multi-modal data fusion; attention mechanisms framework; adaptive decision strategy; predictive performance evaluation; real-time business applications. DOI: 10.1504/IJDMB.2026.10081099
Abstract: Neural performance monitoring in competitive athletes is challenging due to the dynamic and nonlinear nature of physiological and neural indicators. This paper introduces the adaptive gradient dynamics monitor (AGDM), a deep learning framework designed to improve real-time analysis and prediction of athletic performance. AGDM integrates three modules: a convolutional performance tracker for spatiotemporal feature extraction, an agent-driven event segmenter for identifying critical performance events, and an uncertainty-aware optimisation filter for robust predictions. A stochastic refinement scheduling and episodic uncertainty propagation strategy further enhances adaptability by dynamically adjusting learning and uncertainty estimation across temporal sequences. Experimental results demonstrate improved prediction accuracy and robustness compared with traditional methods. The proposed framework provides a scalable solution for neural performance monitoring, supporting enhanced decision-making and personalised training strategies across diverse sports disciplines. Keywords: adaptive gradient dynamics monitor; AGDM; convolutional performance tracker; agent-driven event segmenter; uncertainty-aware optimisation filter; neural performance monitoring; competitive sports. DOI: 10.1504/IJDMB.2026.10081816 Special Issue on: OA Artificial Intelligence for Biomedical, Health Service, and Public Health Knowledge Discovery Abstract: Ethnic vocal music for health promotion carries rich cultural and historical significance, but its composition and preservation face challenges in the digital era. Deep learning offers new possibilities for automatic music generation; however, existing models often lack cultural adaptability and struggle to balance sound quality, style consistency, and emotional expression. This study proposes a hybrid deep learning model combining long short term memory networks and generative adversarial networks for ethnic vocal music for health promotion composition and sound quality optimisation. The LSTM module captures long range temporal dependencies and melodic structures, while the GAN module enhances audio fidelity through adversarial training. An emotion aware loss term and cultural style conditioning vectors are explicitly incorporated to improve emotional expression accuracy and stylistic authenticity. These findings demonstrate that integrating LSTM, GAN, and emotion constraints effectively enhances the quality, authenticity, and cultural relevance of AI generated ethnic vocal music for health promotion. Keywords: long short-term memory; LSTM; generative adversarial network; GAN; ethnic vocal music for health promotion; deep learning; music generation. DOI: 10.1504/IJDMB.2026.10080802
Abstract: Traditional psychological assessment of children’s drawings in preschool children relies heavily on manual feature extraction, resulting in strong subjectivity and inconsistent evaluation standards. Existing deep learning models also encounter difficulty in capturing hidden psychological correlation features embedded in drawings. In addition, conventional optimisation algorithms are prone to falling into local optima during hyperparameter optimisation, which limits assessment accuracy. To address these limitations, an intelligent assessment model integrating improved adaptive particle swarm optimisation (IAPSO) and a convolutional bidirectional attention network (CBAN) is constructed. The proposed tool does not replace the professional judgement of therapists or directly evaluate the effectiveness of therapeutic interventions. Instead, it analyses the features of children’s drawing images and outputs standardised psychological state classification references, thereby providing quantitative support for state monitoring during the art therapy process. Keywords: psychological assessment of children’s drawings; art therapy; deep learning; preschool children’s mental health. DOI: 10.1504/IJDMB.2026.10080803
Abstract: In the existing systems based on cloud computing and virtual reality (VR), playing data may require multiple interactions with remote cloud servers, thus increasing communication delay. Edge computing (EC) migrates some computing and storage tasks from centralised cloud to nodes closer to users. This study focuses on the design of a real-time interactive music therapy assistant system integrating edge computing and deep learning (DL), and takes the improved YOLOv3 model as the edge recognition component. The system collects users' performance data through sensors, and performs local preprocessing and model reasoning at edge nodes. The results show that compared with the contrast system based on DBN, the proposed system can evaluate performance information with higher recognition accuracy and provide feedback with lower delay and resource occupation. These results support the technical feasibility and interactive usability of the system. Keywords: edge computing; EC; music therapy system; real-time interaction; intelligent assistance; artificial intelligence. DOI: 10.1504/IJDMB.2026.10081277
Abstract: To address the deficiencies in existing research, including the absence of an artificial intelligence (AI)-adapted evaluation system for college students’ ideological leadership and weak empirical support, this study adopts a mixed-methods approach. It integrates structural equation modelling (SEM), partial least squares structural equation modelling (PLS-SEM), deep convolutional neural network (DCNN), random forest (RF), support vector machine (SVM), and grounded theory, conducting analysis based on homologous de-sensitised multi-source student data from the authors’ institution. The dataset is stratified and subjected to comprehensive statistical testing. This study clearly defines core concepts, conducts ablation experiments to validate the contributions of individual modules, refines ethical protection protocols for sensitive data, constructs a ‘technology-interaction-value internalisation’ mechanism, and proposes a collaborative optimisation pathway. The research sample is subject to geographical limitations, which warrant expansion and refinement in subsequent studies. Keywords: artificial intelligence; deep learning; ideological leadership; college students; random forest; structural equation modelling; SEM. DOI: 10.1504/IJDMB.2026.10081278
Abstract: With the swift development of mobile games and online multiplayer games, the in-game purchase behaviour of players has become a core component of game revenue. Accurate prediction of players’ purchase behaviour can help operators optimise marketing strategies; it also improves user experience and player retention. To this end, the study proposes a prediction model of in-game purchase behaviour based on the evolving graph convolutional network (EGCN). The model first fuses the category features, temporal behaviours, and continuous features of players; it also extracts key behaviour patterns through an embedding layer and one-dimensional convolution. Then, EGCN is constructed to capture the dynamic changes in players’ social relations over time; the multi-classification prediction of future in-game purchase behaviour is realised by using a gating mechanism and spatial-temporal attention to adaptively focus on key nodes and time points. Keywords: player behaviour prediction; in-game purchase; evolving graph convolution; high-dimensional features; top-K hit rate. DOI: 10.1504/IJDMB.2026.10081279
Abstract: Generative artificial intelligence (AI) offers new possibilities for visual communication design in new media environments, but its application often suffers from uncontrolled generation processes and unclear human-AI role boundaries. This study proposes a structured framework for generative AI-assisted visual communication design specifically for new media tasks. The framework organises the design process into requirement analysis, prompt engineering, multi-round generation, human screening and integration, and publication feedback. A task decomposition mechanism clarifies the functional roles of generative AI and human designers across different design stages, aiming to balance generation efficiency with design controllability. Ablation experiments confirm the contribution of each framework component. This framework provides an operational methodology for human-AI collaborative visual design in new media contexts. Keywords: generative artificial intelligence; visual communication design; new media; human-AI collaboration; prompt engineering. DOI: 10.1504/IJDMB.2026.10081341
Abstract: Aging community renewal requires multi-objective spatial optimisation balancing accessibility, safety, sociability, and comfort. Existing methods rely on static empirical rules and struggle with complex spatial relationships. This study proposes a hybrid graph convolutional network (GCN) framework integrating relational GCN, graph attention, and gated recurrent units. A heterogeneous community graph is constructed to encode physical, social, and functional interactions. An adaptive multi-objective loss dynamically balances conflicting goals. Evaluated on two real communities, the model improves social space utilisation by 61.4% in old neighbourhoods and achieves 92.7 accessibility in new communities. It outperforms baseline GCNs and traditional planning in comprehensive scores. The proposed method provides a data-driven tool for intelligent aging community renewal. Keywords: aging community renewal; spatial layout optimisation; graph convolutional network; multi-objective optimisation; elderly-friendly design; graph attention network. DOI: 10.1504/IJDMB.2026.10081342
Abstract: Music students increasingly face a labour market that expects them to become potential music entrepreneurs. However, existing entrepreneurship and AI-facilitated leadership education often fails to connect their core artistic strengths with entrepreneurial mindsets. This study explored the relationship between three AI-facilitated leadership-related psychological dimensions risk-taking, innovativeness, and proactivity and Chinese music students perceptions of entrepreneurship education. Using a cross-sectional survey design, 196 undergraduate music students from four universities, who had completed a 14-day entrepreneurship education program, were surveyed. Exploratory factor analysis supported a four-factor structure (risk-taking, entrepreneurship education, innovativeness, and proactivity), with factor loadings ranging from approximately 0.52 to 0.83. All subscales demonstrated acceptable internal consistency. The results highlight the need to design AI-facilitated leadership development curricula within music entrepreneurship education, intentionally fostering students risk-taking, proactivity, and innovativeness while explicitly linking these capabilities to their artistic strengths and future career trajectories. Keywords: AI-facilitated leadership; entrepreneurship education; risk-taking; proactivity; music majors. DOI: 10.1504/IJDMB.2026.10081343
Abstract: Against the backdrop of global cultural integration, piano performers face unique psychological challenges in multicultural performance scenarios. This study constructs a competency model for performers’ psychological quality and conducts an empirical analysis of 40 pianists via questionnaires and behavioural observations. Using the global differentiation index and statistical tests, it systematically explores relationships between multicultural elements, psychological pressure, and cultural sensitivity. Results reveal that multicultural contexts significantly increase performers’ psychological load; cultural sensitivity exerts a strong negative predictive effect on stress, with highly sensitive individuals maintaining more stable performance states. Females demonstrate more detailed perceptions of multicultural performance experiences. These findings provide empirical evidence for psychological training and cross-cultural competence development in piano performance, offering practical insights for performing arts education. Keywords: multicultural performance; piano performer; psychological quality; cultural sensitivity; competency model; psychological stress; performing arts education. DOI: 10.1504/IJDMB.2026.10081552
Abstract: This study aims to improve the quality of artificial intelligence-based lyrical soprano singing generation. It investigates the limitations of existing singing synthesis models under challenging conditions, including long musical phrases, register transitions, and sustained high notes, where issues such as rhythmic drift, pitch instability, insufficient lyric intelligibility, unclear phoneme boundaries, and unstable control of vibrato and resonance characteristics commonly occur. This study develops a lyrical soprano generation and optimization framework that integrates audio, musical scores, lyrics, and meta-information. Experimental results demonstrate that the proposed model achieves the lowest average beat deviation of 24.37 and the highest beat alignment accuracy of 97.18% on Subset A. In high-register sections, the F0 root mean square error and semitone pitch error rate are 22.43 and 0.13, respectively. Across the three subsets, the average lyric recognition accuracy, average pronunciation clarity score, and long-term energy smoothness reach 95.30%, 4.52, and 0.048, respectively. Keywords: generation of lyric soprano; multimodal feature fusion; preference alignment; rhythm and pitch recognition. DOI: 10.1504/IJDMB.2026.10081553
Abstract: This paper proposes an integrated framework for personalised physical education that combines dynamic learner modelling with adaptive instructional strategies. DynaLearn-Net employs a dynamic embedding architecture to represent physical, cognitive, and socio-emotional competencies, capturing learner development through competency encoding, task influence networks, graph-based social propagation, and transformer-based temporal updating. Building on this model, EduGameFlow provides real-time adaptive learning strategies through personalised task scheduling, challenge-skill alignment, and curriculum constraints. The framework incorporates peer-matching optimisation, engagement monitoring, and semantic feedback to enhance motivation, collaboration, and instructional responsiveness. By integrating dynamic competency tracking with adaptive educational planning, the proposed approach supports personalised learning experiences and holistic student development. Experimental evaluations demonstrate improvements in learner engagement, developmental progress, and adaptability across diverse educational contexts. These findings highlight the potential of DynaLearn-Net and EduGameFlow to deliver scalable, interpretable, and data-driven solutions for personalised physical education. Keywords: health and exercise; personalised learning; DynaLearn-Net; EduGameFlow; competency modelling; engagement analysis; task scheduling; transformer architecture. DOI: 10.1504/IJDMB.2026.10081679
Abstract: In global learning contexts, intercultural teaching in English as a Foreign Language (EFL) classrooms should not only focus on cultural knowledge comprehension and communicative strategy training. It should also consider whether learners develop healthy interaction states in intercultural communication, such as respect, trust, low-anxiety expression, and psychological safety. This study defines healthy intercultural communicative competence (HICC) as learners' ability in intercultural communication. In this process, learners integrate linguistic, visual, auditory, bodily, and digital interaction resources. This integration supports cultural understanding, positive responses, emotional stability, and constructive negotiation. The changes in HICC-related behaviours were consistent with the decreasing trends in communication apprehension and fear of negative evaluation, indicating an educational spillover effect associated with psychological well-being. This study offers empirical evidence for integrating intercultural communication, psychological safety, and positive social interaction in EFL classrooms. Keywords: multimodal pedagogy; healthy intercultural communicative competence; HICC; EFL classroom; psychological safety; classroom communication behaviours. DOI: 10.1504/IJDMB.2026.10081680 Special Issue on: OA Computational Intelligence for Healthcare Innovations in Data-Driven Diagnosis, Treatment, and Personalised Medicine
Abstract: Traditional health physical fitness assessment is static and relies on single indicators. To enable real-time dynamic analysis, we propose a multi sensor fusion method that integrates physiological and fitness test data. After collection, we apply SVM (optimised by chaotic firefly algorithm) and D-S evidence theory for decision level fusion, and construct a dynamic state space model to track fitness evolution. Experiments on freshmen to seniors at East China Normal University and Shanghai Jiao Tong University School of Medicine, using national physical health standards, measure BMI, cardiopulmonary endurance, lower limb strength, flexibility, and strength. Results show that normal university students’ fitness and cardiopulmonary function decline progressively with grade (male BMI +6.5, female +1.7), while medical university students remain stable due to better diet and exercise habits. The proposed system enables intuitive, real-time fitness analysis, helping schools identify weaknesses and design targeted interventions. Keywords: health physical fitness; physical fitness dynamics; multi sensor fusion; health testing; mathematical technology. DOI: 10.1504/IJDMB.2026.10081346 Special Issue on: OA Hyperautomation and Big Data Methodologies Interactivity and Applications
Abstract: The existing traditional rehabilitation training methods have complex signal processing, low decoding accuracy, and low training efficiency. Therefore, this paper uses deep reinforcement learning (DRL) algorithm to optimise the brain computer interface (BCI) system and optimise and improve the sports rehabilitation training method. Firstly, this paper introduces the current application status of brain computer interface technology in sports rehabilitation and points out the limitations of traditional methods. Then, this paper explains the BCI system implemented using deep reinforcement learning and highlights the benefits of using deep neural networks and proximal strategy optimisation techniques to collect and process patient electroencephalogram (EEG) data. According to the experimental results, compared with traditional methods, the BCI system based on deep reinforcement learning significantly improves training effectiveness and patient engagement, and shortens rehabilitation and recovery time by 4-5 days. Keywords: brain computer interface; BCI; deep reinforcement learning; DRL; sports rehabilitation training; signal processing; personalised rehabilitation; training effect. DOI: 10.1504/IJDMB.2026.10080328
Abstract: This paper proposes an integrated framework for reference-guided animation video colourisation, combining a stylised multimodal transformer (SMT) with style-aware contextual learning (SACL). By jointly modelling reference appearance, target structure, region-level correspondence, and temporal dependencies, the framework improves colour transfer accuracy, perceptual fidelity, and temporal consistency. Experiments on AnimeRun and ATD-12K demonstrate superior performance over existing methods in PSNR, SSIM, LPIPS, and temporal warping error, highlighting its robustness and practical potential. Keywords: deep learning; multimodal transformer; style-aware learning; visual content generation; computational creativity. DOI: 10.1504/IJDMB.2026.10080808
Abstract: Existing image restoration methods rely on a single model, which is difficult to process. Therefore, this study proposes a fused multi-level hybrid restoration framework. Firstly, this paper designs a multi-stage transformer using the Restormer framework to handle compression loss and colour distortion in stages, in order to avoid error accumulation. Then, U-Net semantic segmentation and partial convolution are used to connect high-precision features through skip connections, accurately locate defect areas, and dynamically update mask weights. Finally, this paper uses a diffusion model (denoising diffusion probability model) to gradually generate high fidelity details while preserving the animation art style. The experimental results show that a peak signal-to-noise ratio (PSNR) of 29.45 dB and a structural similarity index of 0.85 were achieved on the ATD-12K dataset. In the case of low compression quality (Q = 10), the PSNR value remained at 21.45 dB. Keywords: advanced algorithm; graphics and image processing; image restoration; compression artefact; restormer framework. DOI: 10.1504/IJDMB.2026.10081344
Abstract: The existing movie restoration methods based on conditional generative adversarial networks (cGAN) lack collaborative modelling mechanisms for structural constraints and colour generation, resulting in structural distortion and colour drift in the restoration results. This study introduces Wasserstein distance into the cGAN framework and constructs a conditional Wasserstein generative adversarial network. Firstly, the degraded film frames are converted into lab space, and the L channel is used as the conditional input for the U-Net encoder. Secondly, the discriminator uses a patch structure combined with Wasserstein distance to constrain the authenticity of the chromaticity distribution. Finally, during the training process, L1 reconstruction, VGG-16 perception, and gradient consistency loss are jointly optimised. The experimental results show that the peak signal-to-noise ratio of this method reaches 26.45 and 25.78 dB on the MPI Sintel and DAVIS test sets, respectively. Keywords: generative adversarial network method; intelligent film restoration; colour enhancement method; structural consistency modelling; multiobjective joint optimisation. DOI: 10.1504/IJDMB.2026.10081345
Abstract: Fine-grained few-shot image classification (FGFSIC) aims to recognize fine-grained categories with limited labeled samples. It remains challenging due to high intra-class variation and low inter-class variation under data scarcity. Existing attention-based methods often fail to capture subtle discriminative patterns and are sensitive to pose variations. To address these issues, this paper proposes a foreground-aware hierarchical relational alignment network (FHRNet), which enhances feature discriminability through cross-level feature interaction and prototype-guided relational modeling. Specifically, a prototype-guided hierarchical relational alignment module hierarchically models region-level correspondences between query features and support prototypes, mitigating the impact of pose and spatial misalignment. Extensive experiments on CUB-200-2011, Stanford-Cars, and Stanford-Dogs demonstrate its effectiveness. With Conv64, FHRNet achieves state-of-the-art performance on CUB-200-2011 and Stanford-Cars, surpassing the best competitor by 5.1% in 1-shot accuracy on CUB-200-2011, while incurring remarkably low computational cost. Keywords: few-shot learning; fine-grained image classification; cross-level feature interaction; prototype-guided hierarchical relational alignment. DOI: 10.1504/IJDMB.2026.10081707
Abstract: The hierarchical feature extraction model with syntactic awareness integrates computer vision and natural language processing to enhance multimodal understanding. Initially, patch partitioning and convolutional layers extract low-level visual features, including edges, textures, and shapes. Two complementary pathways capture global contextual information and fine-grained local details. Hierarchical feature fusion blocks combine these representations to generate coherent multiscale features. Simultaneously, a syntactic mapping layer models grammatical relationships through directed graphs, adjacency functions, and syntactic relation matrices. By aggregating information from neighbouring words, the model enriches word embeddings with semantic and syntactic context. Integrating multiscale visual features with syntax-aware linguistic embeddings produces robust multimodal representations, improving contextual understanding and feature discrimination. The proposed framework supports applications such as image captioning, visual question answering, and cross-modal retrieval, providing a unified approach to visual perception and structured linguistic reasoning. Keywords: multimodal attention; personalised learning; special education; adaptive strategy; learner profiles; resource matching. DOI: 10.1504/IJDMB.2026.10082093
Abstract: Efficient and privacy preserving intelligence at the edge requires chip architectures that can process multimodal data under strict energy and communication constraints. To address this need, we propose a storage computing integrated chip architecture for federated edge intelligence. The framework combines local multimodal learning with privacy aware coordination, thereby reducing data movement while avoiding the transmission of raw user data. At the algorithmic level, the Circuit Sparse Inference Module (CSIM) performs robust sparse representation learning under partial observation through spectral initialisation, hard thresholded encoding, and hardware aware dictionary refinement. The Adaptive Knowledge Embedding Strategy (AKES) further improves representation quality by incorporating feature guided weighting, graph based support refinement, and quantisation aware correction for low bit on chip deployment. Experiments on EdgeBench, TinyImageNet, MobiAct, and SECOM show that the proposed method improves accuracy and convergence while reducing energy consumption and maintaining structural sparsity. These results demonstrate that storage computing integration and federated optimisation can jointly support scalable, energy efficient, and privacy preserving edge intelligence. Keywords: edge computing; federated learning; chip architecture; energy efficiency; privacy preservation. DOI: 10.1504/IJDMB.2026.10082094
Abstract: The current identification of causal effects in e-commerce marketing is inaccurate, making it difficult to provide attribution basis for hyper automated marketing. Therefore, this paper aims to establish an attribution analysis model for the effectiveness of e-commerce marketing activities based on causal reasoning. Firstly, this paper uses dual machine learning (DML) to orthogonally residual high-dimensional user characteristics and marketing intervention variables, weakening the interference of confounding factors. Then, this paper combines causal forest learning to evaluate the effectiveness of user heterogeneity processing, constructs an individual causal contribution estimation model, and combines counterfactual reasoning to calculate incremental contributions and establish a causal attribution matrix for activities. Finally, this paper applies Bayesian posterior inference (BPI) to calibrate the credibility and consistency of attribution results, forming a marketing effectiveness attribution analysis model. The results showed that PEHE reached 0.186 and the estimated deviation of average treatment effect reached 0.0135. Keywords: causal inference; electronic commerce marketing; double machine learning; DML; causal forest; effect attribution. DOI: 10.1504/IJDMB.2026.10082134 |
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:
-
-
