Big Data最新文献

{"title":"Cross-Lingual Real-Time Speech Interaction for Embodied Intelligence: Architecture, Challenges, and Prospects.","authors":"Shuai Wang, Guanzhe Jiao, Yicheng Li","doi":"10.1177/2167647X261487329","DOIUrl":"https://doi.org/10.1177/2167647X261487329","url":null,"abstract":"<p><p>Embodied intelligence is rapidly evolving toward natural human-machine interaction, where real-time cross-lingual speech processing is essential for robots to understand and execute commands across languages. This study reviews research progress from January 2010 to November 1, 2025, using CiteSpace to analyze knowledge structure, hotspots, and trends. The year 2010 was selected as the start date because it marks the early emergence of foundational theoretical discussions on embodied intelligence internationally, providing a comprehensive baseline to observe the entire evolutionary trajectory of the field, even though substantive domestic research emerged later around 2015. Results show a sharp rise in research interest after 2023, driven by multilingual interaction and robotics applications. Current hotspots include multimodal representation learning, low-latency speech-to-text models, cross-lingual semantic alignment, and embodied control. The field has progressed through three stages: conceptual foundations, real-time neural integration, and large-model-driven general embodiment. We summarize the technical framework spanning speech recognition, semantic grounding, and action execution, and identify challenges such as latency reduction, noise robustness, and cross-lingual knowledge transfer. Future work should focus on unified speech-action representation, scalable low-latency architectures, adaptive learning, and ethical governance. This study outlines development trends and prospects for cross-lingual real-time speech interaction in embodied intelligence.</p>","PeriodicalId":51314,"journal":{"name":"Big Data","volume":" ","pages":"2167647X261487329"},"PeriodicalIF":2.0,"publicationDate":"2026-10-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149128665","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"An Edge-Enabled Low-Latency Cross-Lingual Speech-to-Text Framework for Efficient Human-Robot Interaction.","authors":"Abdullah Albuali, Nishant Tripathi, Mahesh T R, Ahlam Almusharraf, Muhammad Faisal Abrar, Isma Farah Siddiqui","doi":"10.1177/2167647X261438099","DOIUrl":"10.1177/2167647X261438099","url":null,"abstract":"<p><p>Humanoid robots are being introduced in places where people do not speak the same language and people expect quick, natural responses. In such situations, speech interaction cannot afford noticeable delays. Most of the present speech-to-text systems are mainly maintained with cloud servers, leading to latency problems, reliance on dependable connectivity, and failures on the fly when used in real time. These shortcomings become especially evident when robots are expected to autonomously and continuously interact with human users. To address these limitations, this project proposes a new edge-centric speech-to-text framework tailored specifically for the multilingual humanoid robot domain. Instead of sending audio data to the cloud, this method performs speech processing directly within the robot. This technology includes lightweight neural models for real-time streaming, an onboard mechanism that allows for real-time identification of the target language, and local caching methods for quicker retrieval of repeated or known speech patterns. Combine these and you can get quicker, more trustworthy transcription without burning a hole in network resources. The system reduces communication delays to a great extent, while providing transcribed data in multiple languages due to local handling of speech from the wireless edge network. The time taken in overall response is more than 60% lower than the response time used in cloud-based systems; it has been found in experiments. More critically, the framework does a good job with fluctuating network bandwidth, loss of packets, and background noise. It is concluded that edge-based and multilingual speech-to-text systems will be important for humanoid robots to enhance responsivity and contextuality. Understanding faster results in faster reactions, smoother conversations, and moments of interaction that feel more natural is a major step toward pragmatic and reliable communication between humans and robots in the working world.</p>","PeriodicalId":51314,"journal":{"name":"Big Data","volume":" ","pages":"336-349"},"PeriodicalIF":2.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147984207","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"TS-PET: A Novel Framework for Fine-Tuning Pretrained Time-Series Models.","authors":"Liyang Zheng, Baisuo Jin","doi":"10.1177/2167647X261439005","DOIUrl":"10.1177/2167647X261439005","url":null,"abstract":"<p><p>Time series foundation models offer powerful zero-shot capabilities but face significant adaptation challenges: (1) inefficient fine-tuning due to computational constraints and unreliable sensitivity scores in existing variants to the low-rank adaptation (LoRA) method, and (2) parameter-heavy prediction heads causing overfitting. We propose time series parameter-efficient transformer (TS-PET), a novel fine-tuning framework featuring: (1) A lightweight prediction module that reduces parameters by >80%, mitigating overfitting while maintaining performance; (2) Specialized pruned LoRA, which introduces robust rank allocation via identifying synergistic parameter interactions, enabling stochastic approximation for efficiency. Extensive experiments on eight benchmarks (Electricity Transforming Temperature, Exchange, Weather, etc.) show TS-PET achieves state-of-the-art accuracy-outperforming MOMENT (linear probing), PatchTST, and adaptive LoRA variants-while demonstrating superior parameter efficiency. Our solution enables scalable adaptation of time series foundation models without compromising performance.</p>","PeriodicalId":51314,"journal":{"name":"Big Data","volume":" ","pages":"323-335"},"PeriodicalIF":2.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147788653","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"ThinkAI: A Natural Language Processing-based Intelligent framework for Mental Health.","authors":"Kashish Ara Shakil, Mudasir Ahmad Wani, Faiz Ullah, Younhyun Jung, Shakir Khan, Gufran Ahmad Ansari","doi":"10.1177/2167647X261438100","DOIUrl":"10.1177/2167647X261438100","url":null,"abstract":"<p><p>In the current digital era, emotional and mental health challenges are becoming very common. Therefore, it is essential to find new and effective ways to support emotional well-being. In this work, we propose ThinkAI helps individuals to better understand and manage their mental health. This platform offers a secure, explainable, and private space where users can express their thoughts and feelings through journaling. ThinkAI then leverages the help of natural language processing (NLP)-based algorithms to analyze these writings to identify emotional patterns. It tracks the changes in the mental well-being of individuals over time. It also offers visual insights that help users see how their emotional states have evolved during that time period, encouraging reflection, and self-awareness. The proposed system also generates alerts in case of adverse situations. ThinkAI comprises two main components: one focused on detecting early signs of depression, and the other on analyzing emotions. The system uses a combination of classical machine learning methods and modern transformer-based models to achieve accurate and reliable results. Both kinds of models were tested. For traditional models, the Support Vector Machine model reported the highest accuracy (0.920) and Receiver Operating Characteristic - Area Under the Curve (ROC-AUC) (0.97) for depression detection, while Naive Bayes had the best recall (0.947). For emotion analysis, Bidirectional Encoder Representations from Transformers (BERT) performed best with an accuracy of 0.945 and an F1-score of 0.9446, closely followed by robustly optimized BERT approach (RoBERTa) and Distilled Robustly Optimized BERT Approach (DistilRoBERTa). Furthermore, to provide a more comprehensive evaluation of the proposed models, we analyzed the training and validation loss across all models up to five epochs, in addition to reporting accuracy. The results highlight how combining classic algorithms with modern transformer-based deep learning models can create powerful tools for understanding emotional and mental health. Thus, ThinkAI offers a promising step toward real-time monitoring of mental health. This work contributes a technically validated and ethically grounded framework. It can be used for real-time monitoring of an individual's mental well-being, digital therapeutics, and large-scale psychological data analysis.</p>","PeriodicalId":51314,"journal":{"name":"Big Data","volume":" ","pages":"350-378"},"PeriodicalIF":2.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147845868","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Neural Gas Network-Based Distinction Between Brain Cysts and Tumors from Magnetic Resonance Images Using a Hybrid Deep Learning Algorithm.","authors":"Mete Yağanoğlu, Öznur Özaltın, Orhan Coşkun","doi":"10.1177/2167647X261490062","DOIUrl":"https://doi.org/10.1177/2167647X261490062","url":null,"abstract":"<p><p>The early and accurate diagnosis of brain tumors is critically important, as timely intervention significantly reduces mortality and improves patient outcomes. While magnetic resonance imaging (MRI) is the preferred diagnostic tool, differentiating malignant brain tumors from benign cysts remains a clinical challenge due to overlapping visual features. This study introduces a technically novel hybrid deep learning framework that integrates denoising, segmentation, and classification stages in a unified pipeline. First, a neural gas network is employed for precise segmentation of MRI images, enhancing boundary delineation between cystic and tumorous regions. To mitigate class imbalance and improve generalization, the Synthetic Minority Oversampling Technique is applied. A unique architectural contribution of this study lies in the integration of three pretrained convolutional neural networks (CNNs)-EfficientNetB0, MobileNetV2, and ResNet50-into a fused hybrid model that combines their complementary feature extraction capabilities. This model is fine-tuned using a grid search-based hyperparameter optimization strategy, early stopping, and cross-validation. Comparative experiments on four distinct datasets (including balanced and segmented versions) demonstrate that the proposed hybrid model consistently outperforms conventional models such as VGG16, InceptionV3, and individual CNN baselines. Achieving a peak accuracy of 97% on the balanced, augmented, and segmented dataset, this work highlights a novel and robust Artificial Intelligence (AI)-based strategy for distinguishing brain cysts from tumors in MRI, with superior performance in complex diagnostic scenarios.</p>","PeriodicalId":51314,"journal":{"name":"Big Data","volume":" ","pages":"2167647X261490062"},"PeriodicalIF":2.0,"publicationDate":"2026-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149065779","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Big Data-Driven Production Scheduling Risk Identification Using a Lightweight Adaptive Graph Attention Network.","authors":"Peng Dong, Ge Han, Luwen Yuan","doi":"10.1177/2167647X261487341","DOIUrl":"https://doi.org/10.1177/2167647X261487341","url":null,"abstract":"<p><p>The advancement of Industry 4.0 and smart manufacturing has led to the generation of high-volume, multisource data that evolve in real-time within production scheduling systems. This poses significant challenges to traditional risk identification methods. Models that ignore topological relationships lack robustness when handling complex dependencies. Black-box deep learning models offer insufficient interpretability to support decision-making. Furthermore, complex graph neural networks entail substantial computational costs and cannot meet real-time requirements. To address these issues, this article proposes an attention-based lightweight graph neural network (AGNN) for big data analytics. The model incorporates a context-aware dynamic attention mechanism that adaptively adjusts node weights based on the global state, thereby enhancing robustness against uncertainties in the scheduling environment. The visuali<i>z</i>ation of attention weights, combined with a business key performance indicator-aligned risk labeling strategy, enhances interpretability and provides a transparent basis for decisions. A lightweight two-layer, single-head architecture maintains high accuracy while compressing the parameter count to 9.8K and achieving an inference speed of 0.163 ms per sample, enabling real-time edge deployment. Importantly, AGNN establishes a data-to-decision link for big data-driven production decision-support systems. It not only identifies risks but also interprets their business impact and propagation pathways, enabling schedulers to prioritize interventions and optimi<i>z</i>e resource allocation in a timely manner. This enhances the adaptability and intelligence of manufacturing systems in dynamic environments. Comparative experiments on a real-world marine shafting dataset demonstrate that AGNN significantly outperforms other baseline models. This study offers a robust, interpretable, and efficient risk identification solution for big data-driven smart manufacturing systems and promotes the application of computational intelligence in complex industrial scenarios within a big data context.</p>","PeriodicalId":51314,"journal":{"name":"Big Data","volume":" ","pages":"2167647X261487341"},"PeriodicalIF":2.0,"publicationDate":"2026-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149003225","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Automated Product Carbon Footprint Estimation via Low-Latency Neural Language Processing in Multilingual, Real-Time Supply Chains.","authors":"Shixiong Guo, Xinzhi Wang","doi":"10.1177/2167647X261481461","DOIUrl":"https://doi.org/10.1177/2167647X261481461","url":null,"abstract":"<p><p>Real-time product carbon footprint (PCF) reporting is increasingly required in global supply chains, yet conventional life cycle assessment (LCA) remains difficult to scale due to fragmented multitier data, heterogeneous formats, and substantial manual effort. This article proposes Auto-LCA, a system-level architecture for continuous, product-level carbon estimation that integrates (i) automated supply-chain data acquisition, (ii) low-latency multilingual natural language processing for extracting life-cycle-relevant attributes from unstructured documents, (iii) a structured product carbon data model, and (iv) machine-learning-based completion of missing life cycle inventory parameters with provenance and confidence tags. A central mechanism is the \"carbon header,\" which packages accumulated embodied and operational emissions as transferable, version-controlled metadata associated with product identifiers and propagated incrementally across supply-chain tiers. Auto-LCA is designed as an open Application Programming Interface (API) layer to interface with enterprise systems (e.g., ERP/PLM/MES) and supports update-driven recomputation that recalculates only affected nodes in the supply-chain graph, improving scalability for large product portfolios and deep supplier networks. Case studies in the steel, cement, and automotive sectors demonstrate that Auto-LCA yields estimates close to representative conventional LCA benchmarks (typically within ±4%), while enabling broader coverage and more frequent updates under realistic data availability constraints. The framework also aligns with emerging digital product passport and sustainability reporting needs by enabling traceable, machine-readable carbon metadata exchange across organizations.</p>","PeriodicalId":51314,"journal":{"name":"Big Data","volume":" ","pages":"2167647X261481461"},"PeriodicalIF":2.0,"publicationDate":"2026-09-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148914190","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Multimodal Graph Neural Networks and Evolutionary Knowledge Fusion for Secure and Explainable Governance in Intelligent IoT Interactive Media Systems.","authors":"Tsung-Chih Hsiao, Tzer-Long Chen","doi":"10.1177/2167647X261481454","DOIUrl":"https://doi.org/10.1177/2167647X261481454","url":null,"abstract":"<p><p>This research proposes a secure, explainable, and context-aware governance framework for blockchain-based digital media contracts in multimodal artificial intelligence-enabled AIoT interactive systems. As digital licensing, NFT copyright management, royalty distribution, and cross-chain content circulation become increasingly embedded in smart media ecosystems, existing contract auditing approaches remain limited by unimodal analysis, weak explainability, black-box decision processes, and insufficient cross-platform generalization. To address these challenges, the proposed framework integrates cross-modal data fusion, graph neural networks, and evolutionary expert knowledge fusion to model smart contract code, abstract syntax trees, control-flow graphs, intercontract transactions, cross-chain records, licensing metadata, NFT copyright attributes, and multimodal IoT interaction logs as unified graph representations. By incorporating causal priors and differentiable rules, the framework supports transparent risk reasoning, verifiable explanations, and trustworthy decision support. The study contributes to explainable blockchain security, multimodal AI governance, and intelligent media systems by enabling more robust detection of copyright misuse, unauthorized licensing, abnormal content distribution, and cross-chain transaction risks.</p>","PeriodicalId":51314,"journal":{"name":"Big Data","volume":" ","pages":"2167647X261481454"},"PeriodicalIF":2.0,"publicationDate":"2026-09-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148876690","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Enhancing Long Short-Term Memory Framework Using Hybrid Meta-Heuristic Optimization Techniques for Traffic Flow Forecasting.","authors":"Rajalakshmi Gurusamy","doi":"10.1177/2167647X261481477","DOIUrl":"https://doi.org/10.1177/2167647X261481477","url":null,"abstract":"<p><p>Short-term traffic flow forecasting is essential for travel safety, congestion avoidance, and effective traffic management as an integral part of intelligent transport systems. Long short-term memory (LSTM) has become a promising technique for forecasting traffic flow. Unfortunately, the LSTM model does not achieve adequate forecast accuracy due to noisy traffic data and poor selection of hyperparameter optimization values. To address these limitations, this study proposes a composite SFFSDF-LSTM model that integrates sinusoidal firefly feature selection (SFFS) and dragonfly (DF) algorithms with an LSTM network. The SFFS algorithm performs adaptive feature selection, while the DF algorithm optimizes hyperparameters to improve convergence and prediction accuracy. The proposed model efficiently identifies the optimal set of LSTM features, weights, biases, and hyperparameters for domain traffic flow prediction while minimizing training errors. From the perspective of error analysis and predictive analytics, the predictive accuracy of the combined model is evaluated and found to be superior to traditional nonparametric and baseline deep learning models.</p>","PeriodicalId":51314,"journal":{"name":"Big Data","volume":" ","pages":"2167647X261481477"},"PeriodicalIF":2.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148867767","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Implicit Bias of Gradient-Based Learning Under Non-Convex Constraints.","authors":"Yanjun Yan","doi":"10.1177/2167647X261465274","DOIUrl":"https://doi.org/10.1177/2167647X261465274","url":null,"abstract":"<p><p>Gradient-based learning under non-convex constraints exhibits a notable phenomenon: Despite the existence of many equivalent global minimizers, optimization algorithms consistently converge to a small subset of structured solutions. This behavior, known as implicit bias, remains insufficiently understood in constrained and non-convex settings. In this article, we investigate the mechanism of implicit bias induced by projected gradient-based optimization over general non-convex feasible sets. By modeling projected gradient descent as a continuous-time dynamical system, we derive a projected gradient flow characterized by tangent and normal cone decompositions, which capture the local geometry of the constraint set. Based on this formulation, we show that constraint geometry continuously filters gradient directions along the optimization trajectory, leading to a trajectory-dependent implicit regularization effect without modifying the objective function. We further formalize this effect through a cumulative normal projection energy functional and prove that the optimization dynamics converge to solutions minimizing both empirical risk and geometric incompatibility with the constraint set. Extensive experiments on synthetic and real-world datasets validate the theoretical predictions, demonstrating consistent alignment between solution geometry, optimization trajectories, and generalization performance. These results provide a unified geometric and dynamical explanation of implicit bias in constrained learning systems.</p>","PeriodicalId":51314,"journal":{"name":"Big Data","volume":" ","pages":"2167647X261465274"},"PeriodicalIF":2.0,"publicationDate":"2026-08-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148842046","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}