Big Data Research最新文献

{"title":"A novel study of kernel graph regularized semi-non-negative matrix factorization with orthogonal subspace for clustering","authors":"Yasong Chen ,&nbsp;Wen Li,&nbsp;Junjian Zhao","doi":"10.1016/j.bdr.2025.100531","DOIUrl":"10.1016/j.bdr.2025.100531","url":null,"abstract":"<div><div>As a nonlinear extension of Non-negative Matrix Factorization (NMF), Kernel Non-negative Matrix Factorization (KNMF) has demonstrated greater effectiveness in revealing latent features from raw data. Building on this, this paper introduces kernel theory and effectively combines the advantages of semi-nonnegative constraints, graph regularization, and orthogonal subspace constraints to propose a novel model-Kernel Graph Regularized Semi-Negative Matrix Factorization with Orthogonal Subspaces and Auxiliary Variables (semi-KGNMFOSV). This model introduces auxiliary variables and reformulates the optimization problem, successfully overcoming the convergence proof challenges typically associated with orthogonal subspace-constrained methods. Furthermore, the model utilizes kernel methods to effectively capture complex nonlinear structures in the data. The semi-nonnegative constraint, along with orthogonal subspace constraints incorporating auxiliary variables, enhances optimization efficiency, while graph regularization preserves the local geometric structure of the data. We develop an efficient optimization algorithm to solve the proposed model and conduct extensive experiments on multiple real-world datasets. Additionally, we investigate the impact of three different initialization strategies on the performance of the proposed algorithm. Experimental results demonstrate that, compared to classical and state-of-the-art methods, the proposed model exhibits superior performance across all three initialization strategies.</div></div>","PeriodicalId":56017,"journal":{"name":"Big Data Research","volume":"40 ","pages":"Article 100531"},"PeriodicalIF":3.5,"publicationDate":"2025-04-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143863357","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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":"TE-PADN: A poisoning attack defense model based on temporal margin samples","authors":"Haitao He ,&nbsp;Ke Liu ,&nbsp;Lei Zhang ,&nbsp;Ke Xu ,&nbsp;Jiazheng Li ,&nbsp;Jiadong Ren","doi":"10.1016/j.bdr.2025.100528","DOIUrl":"10.1016/j.bdr.2025.100528","url":null,"abstract":"<div><div>With the development of network security research, intrusion detection systems based on deep learning show great potential in network attack detection. As crucial tools for ensuring network information security, these systems themselves are vulnerable to poisoning attacks from attackers. Currently, most poisoning attack defense methods cannot effectively utilize network traffic characteristics and are only effective for specific models, showing poor defense results for other models. Furthermore, detection of poisoning attacks is often overlooked, leading to a lack of timely and effective defense against such attacks. Therefore, we propose a data poisoning defense mechanism called TE-PADN. Firstly, we introduce a temporal margin sample generation algorithm that integrates an attention mechanism. Based on mapping the original data time series into a latent feature space, this algorithm learns the temporal characteristics of the data and focuses on information from different positions using the attention mechanism to generate temporal margin samples for repairing poisoned models. Secondly, we propose a multi-level poisoning attack detection method for real-time and accurate detection of undetected poisoning attacks. By employing ensemble learning methods, this approach enhances model robustness, repairs model classification boundaries that have shifted due to poisoning attacks and achieves efficient defense against poisoning attacks. Finally, experimental validation of our proposed method demonstrates promising results. Under a 10% attack intensity, the average accuracy of TE-PADN in recovering poisoning models increased by 6.5% on the NSL-KDD dataset, 5.3% on the UNSW-NB15 dataset, and 5.9% on the CICIDS2017 dataset.</div></div>","PeriodicalId":56017,"journal":{"name":"Big Data Research","volume":"40 ","pages":"Article 100528"},"PeriodicalIF":3.5,"publicationDate":"2025-04-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143816452","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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":"Leveraging artificial intelligence for pandemic management: Case of COVID-19 in the United States","authors":"Ehsan Ahmadi,&nbsp;Reza Maihami","doi":"10.1016/j.bdr.2025.100529","DOIUrl":"10.1016/j.bdr.2025.100529","url":null,"abstract":"<div><div>The COVID-19 pandemic revealed significant limitations in traditional approaches to analyzing time-series data that use one-dimensional data such as historical infection rates. Such approaches do not capture the complex, multifactor influences on disease spread. This paper addresses these challenges by proposing a comprehensive methodology that integrates multiple data sources, including community mobility, census information, Google search trends, socioeconomic variables, vaccination coverage, and political data. In addition, this paper proposes a new cross-learning (CL) methodology that allows for the training of machine learning models on multiple related time series simultaneously, enabling more accurate and robust predictions. Applying the CL approach with four machine learning algorithms, we successfully forecasted confirmed COVID-19 cases 30 days in advance with greater accuracy than the traditional ARIMAX model and the newer Transformer deep learning technique. Our findings identified daily hospital admissions as a significant predictor at the state level and vaccination status at the national level. Random Forest with CL was very effective, performing best in 44 states, while ARIMAX outperformed in seven larger states. These findings highlight the importance of advanced predictive modeling in resource optimization and response strategy development for future health emergencies.</div></div>","PeriodicalId":56017,"journal":{"name":"Big Data Research","volume":"40 ","pages":"Article 100529"},"PeriodicalIF":3.5,"publicationDate":"2025-04-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143839334","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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":"Efficient, interpretable and automated feature engineering for bank data","authors":"Atilla Karaahmetoğlu ,&nbsp;Mehmet Yıldız ,&nbsp;Erdem Ünal ,&nbsp;Uğur Aydın ,&nbsp;Murat Koraş ,&nbsp;Barış Akgün","doi":"10.1016/j.bdr.2025.100524","DOIUrl":"10.1016/j.bdr.2025.100524","url":null,"abstract":"<div><div>Banks rely on expert-generated features and simple models to have high performance and interpretability at the same time. Interpretability is needed for internal assessment and regulatory compliance for specific problems such as risk assessment and both expert generated features and simple models satisfy this need. However, feature generation by experts is a time-consuming process and susceptible to bias. In addition, features need to be generated fairly often due to the dynamic nature of bank data, and in case of significant changes or new data sources, expertise might take a while to build up. Complex models, such as deep neural networks, may be able to remedy this. However, interpretability/explainability approaches for complex models are not satisfactory from the banks' point of view. In addition, such models do not always work well with tabular data which is abundant in banking applications. This paper introduces an automated feature synthesis pipeline that creates informative and domain-interpretable features which iconsumes significantly less time than brute-force methods. We create novel feature synthesis steps, define elimination rules to rule out uninterpretable features, and combine performance-based feature selection methods to pick desirable ones to build our models. Our results on two different datasets show that the features generated with our pipeline; (1) perform on par or better than features generated by existing methods, (2) are obtained faster, and (3) are domain-interpretable.</div></div>","PeriodicalId":56017,"journal":{"name":"Big Data Research","volume":"40 ","pages":"Article 100524"},"PeriodicalIF":3.5,"publicationDate":"2025-03-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143790985","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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":"NoSQL data warehouse optimizing models: A comparative study of column-oriented approaches","authors":"Mohamed Mouhiha,&nbsp;Abdelfettah Mabrouk","doi":"10.1016/j.bdr.2025.100523","DOIUrl":"10.1016/j.bdr.2025.100523","url":null,"abstract":"<div><div>There is a great challenge when building an efficient Big Data Warehouse (DW) from the traditional data warehouse which used to handle the large datasets. Several presented solutions concentrate on the conversion of a standard DW to an columnar model, especially for direct and traditional data sources. Though there have been many successful algorithms that apply data clustering methods, these approaches also come with their fair share of limitations. This paper provides a comprehensive review of the existing methods, both tuned and out-of-the box, exposing their strengths and weaknesses. Further, a comparative study of the different options is always conducted to compare and assess them.</div></div>","PeriodicalId":56017,"journal":{"name":"Big Data Research","volume":"40 ","pages":"Article 100523"},"PeriodicalIF":3.5,"publicationDate":"2025-03-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143681953","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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":"Multi-dimensional feature learning for visible-infrared person re-identification","authors":"Zhenzhen Yang,&nbsp;Xinyi Wu,&nbsp;Yongpeng Yang","doi":"10.1016/j.bdr.2025.100522","DOIUrl":"10.1016/j.bdr.2025.100522","url":null,"abstract":"<div><div>Visible-infrared person re-identification (VI-ReID) is a challenging task due to significant differences between modalities and feature representation of visible and infrared images. The primary goal of current VI-ReID is to reduce discrepancies between modalities. However, existing research primarily focuses on learning modality-invariant features. Due to significant modality differences, it is challenging to learn an effectively common feature space. Moreover, the intra-modality differences have not been well addressed. Therefore, a novel multi-dimensional feature learning network (MFLNet) is proposed in this paper to tackle the inherent challenges of intra-modality and inter-modality differences in VI-ReID. Specifically, to effectively address intra-modality variations, we employ the random local shear (RLS) augmentation, which accurately simulates viewpoint and posture changes. This augmentation can be seamlessly incorporated into other methods without modifying the network or parameters. Additionally, we integrate the multi-dimensional information mining (MIM) module to extract discriminative features and bridge the gap between modalities. Moreover, the cyclical smoothing focal (CSF) loss is introduced to prioritize challenging samples during training, thereby enhancing the ReID performance. Finally, the experimental results indicate that the proposed MFLNet outperforms other VI-ReID approaches on the SYSU-MM01, RegDB and LLCM datasets.</div></div>","PeriodicalId":56017,"journal":{"name":"Big Data Research","volume":"40 ","pages":"Article 100522"},"PeriodicalIF":3.5,"publicationDate":"2025-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143654669","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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":"Deep attention dynamic representation learning networks for recommender system review modeling","authors":"Shivangi Gheewala ,&nbsp;Shuxiang Xu ,&nbsp;Soonja Yeom","doi":"10.1016/j.bdr.2025.100521","DOIUrl":"10.1016/j.bdr.2025.100521","url":null,"abstract":"<div><div>Despite considerable research of utilizing deep learning technology and textual reviews in recommender systems, improving system performance is a contentious matter. This is primarily due to issues faced in learning user-item representations. One issue is the limited ability of networks to model dynamic user-item representations from reviews. Particularly, in sequence-to-sequence learning models, there appears a substantial likelihood of losing semantic knowledge of previous review sequences, as overridden by the next. Another issue lies in effectively integrating global-level and topical-level representations to extract informative content and enhance user-item representations. Existing methods struggle to maintain contextual consistency during this integration process, resulting in suboptimal representation learning, especially attempting to capture finer details. To address these issues, we propose a novel recommendation model called Deep Attention Dynamic Representation Learning (DADRL). Specifically, we employ Latent Dirichlet Allocation and dynamic modulator-based Long Short-Term Memory to extract topical and dynamic global representations. Then, we introduce an attentional fusion methodology to integrate these representations in a contextually consistent manner and construct informative attentional user-item representations. We use these representations into the factorization machines layer to predict the final scores. Experimental results on Amazon categories, Yelp, and LibraryThing show that our model exhibits superior performance compared to several state-of-the-arts. We further examine the DADRL architecture under various conditions to provide insights on the model's employed components.</div></div>","PeriodicalId":56017,"journal":{"name":"Big Data Research","volume":"40 ","pages":"Article 100521"},"PeriodicalIF":3.5,"publicationDate":"2025-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143681952","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Complex data in tourism analysis: A stochastic approach to price competition","authors":"Giovanni Angelini ,&nbsp;Michele Costa ,&nbsp;Andrea Guizzardi","doi":"10.1016/j.bdr.2025.100520","DOIUrl":"10.1016/j.bdr.2025.100520","url":null,"abstract":"<div><div>This study examines pricing strategies and decision-making processes in the hospitality industry by analyzing “ask” prices on online travel agencies (i.e., the rates at which hoteliers are willing to sell their rooms). We face the challenge of modeling a continuous flow of big data organized as “time series of time series,” where daily seasonality and advance bookings intersect. Our research combines insights from tourism, quantitative methods, and big data to improve pricing strategies, contributing to both theory and practice in revenue management. Focusing on Venice, we analyze price competition as a multivariate stochastic process using a Structural Vector Autoregressive (SVAR) approach, aligning with modern dynamic pricing algorithms.</div><div>The findings show that time-based pricing strategies, which adjust based on the day of arrival and booking, are more important than room features in setting hotel prices. We also find that price changes have a non-linear and decreasing effect as the booking date approaches. These insights suggest that hotels could create more advanced pricing strategies, and policymakers should consider these factors when addressing the challenges related to overtourism.</div><div>We study the complex competitive relationships among heterogeneous service providers with an approach applicable to any market where consumption is delayed relative to purchase time. However, we highlight that the quality and accessibility of information in the tourism sector are key aspects to be considered when using big data in this industry.</div></div>","PeriodicalId":56017,"journal":{"name":"Big Data Research","volume":"40 ","pages":"Article 100520"},"PeriodicalIF":3.5,"publicationDate":"2025-03-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143641669","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"ImDMI: Improved Distributed M-Invariance model to achieve privacy continuous big data publishing using Apache Spark","authors":"Salheddine Kabou ,&nbsp;Laid Gasmi ,&nbsp;Abdelbaset Kabou ,&nbsp;Sidi Mohammed Benslimane","doi":"10.1016/j.bdr.2025.100519","DOIUrl":"10.1016/j.bdr.2025.100519","url":null,"abstract":"<div><div>One of the critical challenges in the big data analytics is the individual's privacy issues. Data anonymization models including k-anonymity and l-diversity are used to guarantee the tradeoff between privacy and data utility while publishing the data. However, these models focus only on the single release of datasets and produce a certain level of privacy. In practical big data applications, data publishing is more complicated where the data is published continuously as new data is collected, and the privacy should be achieved for different releases. In this research, we propose a new distributed bottom up approach on Apache Spark for achievement of the m-invariance privacy model in the continuous big data context. The proposed approach, which is the first study that deals with dynamic big data publishing, is based on the insertion and the split process. In the first process, the data records collected from different workers are inserted into an improved bottom up R-tree generalization in order to minimizing the information loss. The second process concentrates on splitting the overflowed node with respect to the m-invariance model requirement by minimizing the overlap between the resulting partitions. The experimental results show significant improvement in term of data utility, execution time and counterfeit data records as compared to existing techniques in the literature.</div></div>","PeriodicalId":56017,"journal":{"name":"Big Data Research","volume":"40 ","pages":"Article 100519"},"PeriodicalIF":3.5,"publicationDate":"2025-03-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143609162","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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":"Predicting option prices: From the Black-Scholes model to machine learning methods","authors":"Angela Maria D'Uggento,&nbsp;Marta Biancardi,&nbsp;Domenico Ciriello","doi":"10.1016/j.bdr.2025.100518","DOIUrl":"10.1016/j.bdr.2025.100518","url":null,"abstract":"<div><div>In the ever-changing landscape of financial markets, accurate option pricing remains critical for investors, traders and financial institutions. Traditionally, the Black-Scholes (B&amp;S) model has been the cornerstone for option pricing, providing a solid framework based on mathematical and physical principles. Nevertheless, the B&amp;S model has some limitations, such as the restriction to European options, the absence of dividends, constant volatility, etc. Studies and academic literature on the application of machine learning models in the financial sector are rapidly increasing. The main objective of this paper is to provide a comprehensive comparative analysis between the traditional B&amp;S model and the most commonly used machine learning algorithms such as Artificial Neural Networks (ANNs). The rationale is twofold. First, to examine the assumptions of the B&amp;S model, such as constant volatility and a perfectly efficient market, in light of the complexity of the real world, even though it is recognized that the model has been known as a pillar for decades. Secondly, to emphasize that the proliferation of big data and advances in computing power have fuelled the rise of machine learning techniques in finance. These algorithms have remarkable capabilities in discovering non-linear patterns and extracting information from large data sets, providing a compelling alternative to traditional quantitative methods. Machine learning offers a new way to capture and model such complex financial dynamics, which can lead to more accurate pricing models. By comparing the B&amp;S model and some machine learning approaches, this paper aims to shed light on their respective strengths, weaknesses and applicability in the context of options pricing using real data. Through rigorous empirical analyses and performance metrics, our results demonstrate the importance of using machine learning techniques that can outperform or complement the established B&amp;S model in predicting option prices by achieving higher prediction accuracy.</div></div>","PeriodicalId":56017,"journal":{"name":"Big Data Research","volume":"40 ","pages":"Article 100518"},"PeriodicalIF":3.5,"publicationDate":"2025-02-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143520058","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}