Big Data Cogn. Comput., Volume 9, Issue 8 (August 2025) – 16 articles

In the highly interconnected digital ecosystem, cyberspace has become the main battlefield for complex attacks such as Advanced Persistent Threat (APT). The complexity and concealment of APT attacks are increasing, posing unprecedented challenges to network security. Current APT detection methods largely depend on general datasets, making it challenging to capture the stages and complexity of APT attacks. Moreover, existing detection methods often suffer from suboptimal accuracy, high false alarm rates, and a lack of real-time capabilities. In this paper, we introduce LDR-RFECV, a novel feature selection (FS) algorithm that uses LightGBM, Decision Trees (DTs), and Random Forest (RF) as integrated feature evaluators instead of single evaluators in recursive feature elimination algorithms. This approach helps select the optimal feature subset, thereby significantly enhancing detection efficiency. In addition, a novel optimization algorithm called LWHO was proposed, which integrates the Levy flight mechanism with the Wild Horse Optimizer (WHO) to optimize the hyperparameters of the LightGBM model, ultimately enhancing performance in APT attack detection. More importantly, this optimization strategy significantly boosts the detection rate during the lateral movement phase of APT attacks, a pivotal stage where attackers infiltrate key resources. Timely identification is essential for disrupting the attack chain and achieving precise defense. Experimental results demonstrate that the proposed method achieves 97.31% and 98.32% accuracy on two typical APT attack datasets, DAPT2020 and Unraveled, respectively, which is 2.86% and 4.02% higher than the current research methods, respectively. Full article

Customers heavily depend on reviews for product information. Fake reviews may influence the perception of product quality, making online reviews less effective. ChatGPT’s (GPT-3.5 and GPT-4) ability to generate human-like reviews and responses to inquiries across several disciplines has increased recently. This leads to an increase in the number of reviewers and applications using ChatGPT to create fake reviews. Consequently, the detection of fake reviews generated or rephrased by ChatGPT has become essential. This paper proposes a new approach that distinguishes ChatGPT-rephrased reviews, considered fake, from real ones, utilizing a balanced dataset to analyze the sentiment and linguistic patterns that characterize both reviews. The proposed model further leverages Explainable Artificial Intelligence (XAI) techniques, including Local Interpretable Model-agnostic Explanations (LIME) and Shapley Additive Explanations (SHAP) for deeper insights into the model’s predictions and the classification logic. The proposed model performs a pre-processing phase that includes part-of-speech (POS) tagging, word lemmatization, tokenization, and then fine-tuned Transformer-based Machine Learning (ML) model DistilBERT for predictions. The obtained experimental results indicate that the proposed fine-tuned DistilBERT, utilizing the constructed balanced dataset along with a pre-processing phase, outperforms other state-of-the-art methods for detecting ChatGPT-rephrased reviews, achieving an accuracy of 97.25% and F1-score of 97.56%. The use of LIME and SHAP techniques not only enhanced the model’s interpretability, but also offered valuable insights into the key factors that affect the differentiation of genuine reviews from ChatGPT-rephrased ones. According to XAI, ChatGPT’s writing style is polite, uses grammatical structure, lacks specific descriptions and information in reviews, uses fancy words, is impersonal, and has deficiencies in emotional expression. These findings emphasize the effectiveness and reliability of the proposed approach. Full article

In real-world visual recognition tasks, long-tailed distribution is a pervasive challenge, where the extreme class imbalance severely limits the representation learning capability of deep models. Although supervised learning has demonstrated certain potential in long-tailed visual recognition, these models’ gradient updates dominated by head classes often lead to insufficient representation of tail classes, resulting in ambiguous decision boundaries. While existing Supervised Contrastive Learning variants mitigate class bias through instance-level similarity comparison, they are still limited by biased negative sample selection and insufficient modeling of the feature space structure. To address this, we propose Rebalancing Supervised Contrastive Learning (Reb-SupCon), which constructs a balanced and discriminative feature space during model training to alleviate performance deviation. Our method consists of two key components: (1) a dynamic rebalancing factor that automatically adjusts sample contributions through differentiable weighting, thereby establishing class-balanced feature representations; (2) a prototype-aware enhancement module that further improves feature discriminability by explicitly constraining the geometric structure of the feature space through introduced feature prototypes, enabling locally discriminative feature reconstruction. This breaks through the limitations of conventional instance contrastive learning and helps the model to identify more reasonable decision boundaries. Experimental results show that this method demonstrates superior performance on mainstream long-tailed benchmark datasets, with ablation studies and feature visualizations validating the modules’ synergistic effects. Full article

Backfilling materials are commonly employed materials in mines for filling mining waste, and the strength of the consolidated backfill formed by the binding material directly influences the stability of the surrounding rock and production safety in mines. The traditional approach to obtaining the strength of the backfill demands a considerable amount of manpower and time. The rapid and precise acquisition and optimization of backfill strength parameters hold utmost significance for mining safety. In this research, the authors carried out a backfill strength experiment with five experimental parameters, namely concentration, cement–sand ratio, waste rock–tailing ratio, curing time, and curing temperature, using an orthogonal design. They collected 174 sets of backfill strength parameters and employed six population optimization algorithms, including the Artificial Ecosystem-based Optimization (AEO) algorithm, Aquila Optimization (AO) algorithm, Germinal Center Optimization (GCO), Sand Cat Swarm Optimization (SCSO), Sparrow Search Algorithm (SSA), and Walrus Optimization Algorithm (WaOA), in combination with the CatBoost algorithm to conduct a prediction study of backfill strength. The study also utilized the Shapley Additive explanatory (SHAP) method to analyze the influence of different parameters on the prediction of backfill strength. The results demonstrate that when the population size was 60, the AEO-CatBoost algorithm model exhibited a favorable fitting effect (R² = 0.947, VAF = 93.614), and the prediction error was minimal (RMSE = 0.606, MAE = 0.465), enabling the accurate and rapid prediction of the strength parameters of the backfill under different ratios and curing conditions. Additionally, an increase in curing temperature and curing time enhanced the strength of the backfill, and the influence of the waste rock–tailing ratio on the strength of the backfill was negative at a curing temperature of 50 °C, which is attributed to the change in the pore structure at the microscopic level leading to macroscopic mechanical alterations. When the curing conditions are adequate and the parameter ratios are reasonable, the smaller the porosity rate in the backfill, the greater the backfill strength will be. This study offers a reliable and accurate method for the rapid acquisition of backfill strength and provides new technical support for the development of filling mining technology. Full article

The Evidential K-Nearest Neighbor (EK-NN) classifier has demonstrated robustness in handling incomplete and uncertain data; however, its application in high-dimensional big data for feature selection, such as genomic datasets with tens of thousands of gene features, remains underexplored. Our proposed Granular–Elastic Evidential K-Nearest Neighbor (GEK-NN) approach addresses this gap. In the context of big data, GEK-NN integrates an Elastic Net within the Genetic Algorithm’s fitness function to efficiently sift through vast amounts of data, identifying relevant feature subsets. This process mimics human cognitive behavior of filtering and refining information, similar to concepts in cognitive computing. A granularity metric is further employed to optimize subset size, maximizing its impact. GEK-NN consists of two crucial phases. Initially, an Elastic Net-based feature evaluation is conducted to pinpoint relevant features from the high-dimensional data. Subsequently, granularity-based optimization refines the subset size, adapting to the complexity of big data. Before applying to genomic big data, experiments on UCI datasets demonstrated the feasibility and effectiveness of GEK-NN. By using an Evidence Theory framework, GEK-NN overcomes feature-selection challenges in both low-dimensional UCI datasets and high-dimensional genomic big data, significantly enhancing pattern recognition and classification accuracy. Comparative analyses with existing EK-NN feature-selection methods, using both UCI and high-dimensional gene datasets, underscore GEK-NN’s superiority in handling big data for feature selection and classification. These results indicate that GEK-NN not only enriches EK-NN applications but also offers a cognitive-inspired solution for complex gene data analysis, effectively tackling high-dimensional feature-selection challenges in the realm of big data. Full article

With the accelerating globalization and rapid development of science and technology, scientific collaboration has become a key driver of knowledge production, yet its patterns vary significantly across disciplines. This study aims to explore the disciplinary differences in scholars’ scientific collaboration patterns and their underlying mechanisms. Data were collected from the China National Knowledge Infrastructure (CNKI) database, covering papers from four disciplines: mathematics, mechanical engineering, philosophy, and sociology. Using social network analysis, we examined core network metrics (degree centrality, neighbor connectivity, clustering coefficient) in collaboration networks, analyzed collaboration patterns across scholars of different academic ages, and compared the academic age distribution of collaborators and network characteristics across career stages. Key findings include the following. (1) Mechanical engineering exhibits the highest and most stable clustering coefficient (mean 0.62) across all academic ages, reflecting tight team collaboration, with degree centrality increasing fastest with academic age (3.2 times higher for senior vs. beginner scholars), driven by its reliance on experimental resources and skill division. (2) Philosophy shows high degree centrality in early career stages (mean 0.38 for beginners) but a sharp decline in clustering coefficient in senior stages (from 0.42 to 0.17), indicating broad early collaboration but loose later ties due to individualized knowledge production. (3) Mathematics scholars prefer collaborating with high-centrality peers (higher neighbor connectivity, mean 0.51), while sociology shows more inclusive collaboration with dispersed partner centrality. Full article

Currently, various pruning strategies including different methods and distribution types are commonly used to reduce the number of FLOPs and parameters in deep learning models. However, their impact on actual energy savings remains insufficiently studied, particularly in resource-constrained settings. To address this, we introduce PruneEnergyAnalyzer, an open-source Python tool designed to evaluate the energy efficiency of pruned models. Starting from the unpruned model, the tool calculates the energy savings achieved by pruned versions provided by the user, and generates comparative visualizations based on previously applied pruning hyperparameters such as method, distribution type (PD), compression ratio (CR), and batch size. These visual outputs enable the identification of the most favorable pruning configurations in terms of FLOPs, parameter count, and energy consumption. As a demonstration, we evaluated the tool with 180 models generated from three architectures, five pruning distributions, three pruning methods, and four batch sizes, using another previous library (e.g. FlexiPrune). This experiment revealed the significant impact of the network architecture on Energy Reduction, the non-linearity between FLOPs savings and energy savings, as well as between parameter reduction and energy efficiency. It also showed that the batch size strongly influences the energy consumption of the pruned model. Therefore, this tool can support researchers in making pruning policy decisions that also take into account the energy efficiency of the pruned model. Full article

This study presents the first systematic empirical comparison of neural architectures for Universal Dependencies (UD) parsing in Old English, thus addressing central questions in computational historical linguistics and low-resource language processing. We evaluate three approaches—a baseline spaCy pipeline, a pipeline with a pretrained tok2vec component, and a MobileBERT transformer-based model—across datasets ranging from 1000 to 20,000 words. Our results demonstrate that the pretrained tok2vec model consistently outperforms alternatives, because it achieves 83.24% UAS and 74.23% LAS with the largest dataset, whereas the transformer-based approach substantially underperforms despite higher computational costs. Performance analysis reveals that basic tagging tasks reach 85–90% accuracy, while dependency parsing achieves approximately 75% accuracy. We identify critical scaling thresholds, with substantial improvements occurring between 1000 and 5000 words and diminishing returns beyond 10,000 words, which provides insights into scaling laws for historical languages. Technical analysis reveals that the poor performance of the transformer stems from parameter-to-data ratio mismatches (1250:1) and the unique orthographic and morphological characteristics of Old English. These findings defy assumptions about transformer superiority in low-resource scenarios and establish evidence-based guidelines for researchers working with historical languages. The broader significance of this study extends to enabling an automated analysis of three million words of extant Old English texts and providing a framework for optimal architecture selection in data-constrained environments. Our results suggest that medium-complexity architectures with monolingual pretraining offer superior cost–benefit trade-offs compared to complex transformer models for historical language processing. Full article

End-to-end relation extraction (E2ERE) generally performs named entity recognition and relation extraction either simultaneously or sequentially. While numerous studies on E2ERE have centered on enhancing span representations to improve model performance, challenges remain due to the gaps between subtasks (named entity recognition and relation extraction) and the modeling discrepancies between entities and relations. In this paper, we propose a novel Label Annotation Interaction-based representation enhancement method for E2ERE, which institutes a two-phase semantic interaction to augment representations. Specifically, we firstly feed label annotations that are easy to manually annotate into a language model, and conduct the first-round interaction between three types of tokens with a partial attention mechanism; Then we construct a latent multi-view graph to capture various possible links between label and entity (pair) nodes, facilitating the second-round interaction between entities and labels. A series of comparative experiments with methods of various transformer-based architectures currently in use show that LAI-Net can maintain performance on par with the current SOTA in terms of NER task, and achieves significant improvements over existing SOTA models in terms of RE task. Full article

With the exponential growth of textual data, traditional topic modeling methods based on static analysis demonstrate limited effectiveness in tracking the dynamics of thematic content. This research aims to develop a method for quantifying the dynamics of topics within text corpora using a thematic signal (TS) function that accounts for temporal changes and semantic relationships. The proposed method combines associative tokens with original lexical units to reduce thematic entropy and information noise. Approaches employed include topic modeling (LDA), vector representations of texts (TF-IDF, Word2Vec), and time series analysis. The method was tested on a corpus of news texts (5000 documents). Results demonstrated robust identification of semantically meaningful thematic clusters. An inverse relationship was observed between the level of thematic significance and semantic diversity, confirming a reduction in entropy using the proposed method. This approach allows for quantifying topic dynamics, filtering noise, and determining the optimal number of clusters. Future applications include analyzing multilingual data and integration with neural network models. The method shows potential for monitoring information flows and predicting thematic trends. Full article

This study evaluates leadership uniformity—the degree to which the proposer role is evenly distributed among validator nodes over time—in Quorum-based Byzantine Fault Tolerance (QBFT), a Byzantine Fault-Tolerant (BFT) consensus algorithm used in permissioned blockchain networks. By introducing simulated follower timeouts derived from uniform, normal, lognormal, and Weibull distributions, it models a range of network conditions and latency patterns across nodes. This approach integrates Raft-inspired timeout mechanisms into the QBFT framework, enabling a more detailed analysis of leader selection under different network conditions. Three leader selection strategies are tested: Direct selection of the node with the shortest timeout, and two quorum-based approaches selecting from the top 20% and 30% of nodes with the shortest timeouts. Simulations were conducted over 200 rounds in a 10-node network. Results show that leader selection was most equitable under the Weibull distribution with shape $k=0.5$, which captures delay behavior observed in real-world networks. In contrast, the uniform distribution did not consistently yield the most balanced outcomes. The findings also highlight the effectiveness of quorum-based selection: While choosing the node with the lowest timeout ensures responsiveness in each round, it does not guarantee uniform leadership over time. In low-variability distributions, certain nodes may be repeatedly selected by chance, as similar timeout values increase the likelihood of the same nodes appearing among the fastest. Incorporating controlled randomness through quorum-based voting improves rotation consistency and promotes fairer leader distribution, especially under heavy-tailed latency conditions. However, expanding the candidate pool beyond 30% (e.g., to 40% or 50%) introduced vote fragmentation, which complicated quorum formation in small networks and led to consensus failure. Overall, the study demonstrates the potential of timeout-aware, quorum-based leader selection as a more adaptive and equitable alternative to round-robin approaches, and provides a foundation for developing more sophisticated QBFT variants tailored to latency-sensitive networks. Full article

Emotion detection using computer vision has advanced significantly in recent years, achieving remarkable performance that, in some cases, surpasses that of humans. Convolutional neural networks (CNNs) excel in this task by capturing facial features that allow for effective emotion classification. However, most research focuses on basic emotions, such as happiness, anger, or sadness, neglecting more complex emotions, like frustration. People set expectations or goals to meet; if they do not happen, frustration arises, generating reactions such as annoyance, anger, and disappointment, which can harm confidence and motivation. These aspects make it especially relevant in mental health and educational contexts, where detecting it could help mitigate its adverse effects. In this research, we developed a CNN-based approach to detect frustration through facial expressions. The scarcity of specific datasets for this task led us to create an experimental protocol to generate our dataset. This classification task presents a high degree of difficulty due to the variability in facial expressions among different participants when feeling frustrated. Despite this, our new model achieved an F1-score of $0.8080$, thus obtaining an adequate baseline model. Full article

Customer sentiment analysis plays a pivotal role in the digital economy by offering comprehensive insights that inform strategic business decisions, optimize digital marketing initiatives, and improve overall customer satisfaction. We propose a large language model-based topic-level sentiment analysis framework. We employ a BERT-based model to generate contextualized vector representations of the documents, and then clustering algorithms are automatically applied to group documents into topics. Once the topics are formed, a GPT model is used to perform sentiment classification on the content related to each topic. The simulations show the effectiveness of this approach, where selecting appropriate clustering techniques yields more semantically coherent topics. Furthermore, topic-level sentiment polarization shows that 31.7% of all negative sentiment concentrates on the shopping experience, despite an overall positive sentiment trend. Full article

The rapid advancement of artificial intelligence (AI) and machine learning has revolutionised how systems process information, make decisions, and adapt to dynamic environments. AI-driven approaches have significantly enhanced efficiency and problem-solving capabilities across various domains, from automated decision-making to knowledge representation and predictive modelling. These developments have led to the emergence of increasingly sophisticated models capable of learning patterns, reasoning over complex data structures, and generalising across tasks. As AI systems become more deeply integrated into networked infrastructures and the Internet of Things (IoT), their ability to process and interpret data in real-time is essential for optimising intelligent communication networks, distributed decision making, and autonomous IoT systems. However, despite these achievements, the internal mechanisms that drive LLMs’ reasoning and generalisation capabilities remain largely unexplored. This lack of transparency, compounded by challenges such as hallucinations, adversarial perturbations, and misaligned human expectations, raises concerns about their safe and beneficial deployment. Understanding the underlying principles governing AI models is crucial for their integration into intelligent network systems, automated decision-making processes, and secure digital infrastructures. This paper provides a comprehensive analysis of explainability approaches aimed at uncovering the fundamental mechanisms of LLMs. We investigate the strategic components contributing to their generalisation abilities, focusing on methods to quantify acquired knowledge and assess its representation within model parameters. Specifically, we examine mechanistic interpretability, probing techniques, and representation engineering as tools to decipher how knowledge is structured, encoded, and retrieved in AI systems. Furthermore, by adopting a mechanistic perspective, we analyse emergent phenomena within training dynamics, particularly memorisation and generalisation, which also play a crucial role in broader AI-driven systems, including adaptive network intelligence, edge computing, and real-time decision-making architectures. Understanding these principles is crucial for bridging the gap between black-box AI models and practical, explainable AI applications, thereby ensuring trust, robustness, and efficiency in language-based and general AI systems. Full article

Linguistic steganography can be utilized to establish covert communication channels on social media platforms, thus facilitating the dissemination of illegal messages, seriously compromising cyberspace security. Synonym substitution-based linguistic steganography methods have garnered considerable attention due to their simplicity and strong imperceptibility. Existing linguistic steganalysis methods have not achieved excellent detection performance for the aforementioned type of linguistic steganography. In this paper, based on the idea of focusing on accumulated differences, we propose a two-stage synonym substitution-based linguistic steganalysis method that does not require a synonym database and can effectively detect texts with very low embedding rates. Experimental results demonstrate that this method achieves an average detection accuracy 2.4% higher than the comparative method. Full article

Face recognition (FR) in unconstrained conditions remains an open research topic and an ongoing challenge. The facial images exhibit diverse expressions, occlusions, variations in illumination, and heterogeneous backgrounds. This work aims to produce an accurate and robust system for enhanced Security and Surveillance. A parallel hybrid deep learning model for feature extraction and classification is proposed. An ensemble of three parallel extraction layer models learns the best representative features using CNN and RNN. 2D LBP and 3D Mesh LBP are computed on face images to extract image features as input to two RNNs. A stacked autoencoder (SAE) merged the feature vectors extracted from the three CNN-RNN parallel layers. We tested the designed 2D/3D CNN-RNN framework on four standard datasets. We achieved an accuracy of $98.9\%$. The hybrid deep learning model significantly improves FR against similar state-of-the-art methods. The proposed model was also tested on an unconstrained conditions human crowd dataset, and the results were very promising with an accuracy of $95\%$. Furthermore, our model shows an 11.5% improvement over similar hybrid CNN-RNN architectures, proving its robustness in complex environments where the face can undergo different transformations. Full article