Annals of Data Science最新文献

With the increasing complexity of oil and gas pipeline networks, early identification of leaks and defects is crucial to ensure the safe operation of pipelines. This study proposes a graph neural network (GNN) method for data processing and defect identification aimed at optimizing monitoring and maintenance strategies for oil and gas pipelines. Through the analysis of historical leakage data, we constructed a graph database containing 5000 samples, each containing 10 features such as pressure, flow, temperature, etc. Using graph convolutional network and graph attention network (GAT) to perform feature extraction and pattern recognition on nodes in pipeline network, our model achieves 92% accuracy in defect recognition, which is 15% higher than traditional methods. In addition, we have developed a leakage prediction model based on time series analysis, which is able to predict potential leakage risks 24 h in advance with an accuracy of 85%. The results of this study not only improve the safety management level of oil and gas pipelines, but also provide a new technical path for future intelligent pipeline maintenance.

This research investigates the application of the Machine Learning (ML) model for effective and equitable essay scoring in education. Unlike their human counterpart, ML models have the capacity to rapidly analyze scores of essays, providing timely and equitable scores that take into account varying student demographics and styles of writing. This function helps in the identification of classroom problems and supports the design of focused teaching methodologies. For the study, a Light Gradient Boosting Classification (LGBC) model was optimized by three optimizers: Black Widow Optimization (BWO), Zebra Optimization Algorithm (ZOA), and Leader Harris Hawks Optimization (LHHO), for the development of the hybrid models with a focus on improved prediction quality. Comparison of these hybrid models with the base LGBC model was performed through different phases, such as Training, Validation, and Testing. The findings show that the LGLH model exhibited improved performance with an accuracy rate of 0.981, followed by the LGZO model with 0.971 and the LGBW model with 0.963. The lowest rate of accuracy was observed in the base LGBC model, which was 0.946. The results demonstrate the efficacy of hybrid models, which harness the optimality of several optimization techniques and provide more robust results for complicated tasks. The study emphasizes the importance of selecting the appropriate model architecture to achieve optimal performance, providing valuable insights into model efficacy at various stages of evaluation.

In university admissions, interaction networks naturally emerge between prospective students and available majors. Understanding hidden patterns in such a vast network is crucial for decision-making but poses technical challenges due to its complexity and data limitations. Many existing models rely heavily on user profiling, raising privacy concerns and making data collection difficult. Instead, this work extracts meaningful insights using only the adjacency information of the network, avoiding the need for personal data. We leverage Graph Convolutional Networks (GCN) to generate compact representations for major recommendation and clustering tasks. Our GCN-based approach outperforms classical methods such as popularity-based and Non-negative Matrix Factorization (NMF), as well as the neural Generalized Matrix Factorization (GMF) model, achieving up to 61.06% and 12.17% improvements in smaller (dimension 40) and larger (dimension 80) embeddings, respectively. Furthermore, hierarchical clustering on these embeddings reveals implicit patterns in student preferences, particularly regarding fields of study and geographic locations, even without explicit data on these attributes. These findings demonstrate that meaningful insights can be derived from interaction networks while mitigating privacy concerns associated with user profiling.

In the rapidly evolving landscape of online information dissemination, managing rumors has become an imperative challenge for governments worldwide. This study employs a tripartite evolutionary game model to examine the behavior evolution of the government, online media, and netizens in the process of rumor propagation under uncertain conditions. The innovation of the model lies in considering the probability of successful rumor detection under government regulation, the uncertainty of rumor dissemination by online media and netizens, and introducing a dynamic government penalty mechanism. Through simulation and analysis, we identify the evolutionarily stable strategies of each participant under different scenarios and provide specific governance strategies for each party involved. The results reveal that appropriate government penalties, proactive regulation by online media, and rational choices by netizens can effectively curb rumor spreading. In uncertain environments, adopting flexible policies and dynamic adjustment mechanisms is crucial for effective rumor governance. The results reveal that appropriate government penalties, proactive regulation by online media, and rational choices by netizens can effectively curb rumor spreading. In uncertain environments, adopting flexible policies and dynamic adjustment mechanisms is crucial for effective rumor governance. This study not only enriches the application of evolutionary game theory but also offers practical strategic recommendations for policymakers to address the challenges of rumor propagation.

Automatic generation of test cases using heuristic methods is a hot research topic nowadays. Although its advantages are obvious, it is slightly insufficient in the selection of optimal individuals. Aiming at the existing problems in the evaluation and selection of the optimal individual, this paper proposes a test case evaluation algorithm based on the comprehensive analysis of the characteristics of layer proximity and branch distance function, which is a joint structure of “layer proximity and branch distance function”. The basic idea of this algorithm is that when selecting pilot individuals in the evolutionary process, we first select the individuals with high proximity between the actual execution path and the target path, and then select the individuals with the smallest branching distances among these individuals, so as to obtain the individuals with the optimal piloting ability. Experiments show that the proposed algorithm can quickly find the optimal test cases, especially for the test case generation of multi-layer nested programs.

Detectors have been extensively utilized in various scenarios such as autonomous driving and video surveillance. Nonetheless, recent studies have revealed that these detectors are vulnerable to adversarial attacks, particularly adversarial patch attacks. Adversarial patches are specifically crafted to disrupt deep learning models by disturbing image regions, thereby misleading the deep learning models when added to into normal images. Traditional adversarial patches often lack semantics, posing challenges in maintaining concealment in physical world scenarios. To tackle this issue, this paper proposes a Prompt-based Natural Adversarial Patch generation method, which creates patches controllable by textual descriptions to ensure flexibility in application. This approach leverages the latest text-to-image generation model—Latent Diffusion Model (LDM) to produce adversarial patches. We optimize the attack performance of the patches by updating the latent variables of LDM through a combined loss function. Experimental results indicate that our method can generate more natural, semantically rich adversarial patches, achieving effective attacks on various detectors.

This study examines the issue of green information distortion and its impact on tourists’ purchasing decisions, as well as the associated high transaction costs within the green tourism supply chain. By selecting a green tourism supply chain with varying government subsidy schemes as the focus of the research, the objective is to explore optimal subsidy strategies and assess the implications of blockchain integration. A three-level Stackelberg game model is established, featuring the government as the leader and a scenic spot (SS) and travel agency (TA) as participants. Key findings include: (1) Production subsidies are more effective in boosting market demand than environmental investment subsidies, particularly when tourist green trust and preferences are high. (2) Blockchain enhances greenness, market demand, and social welfare, positively influencing the green tourism supply chain (GTSC). (3) Tourist green preference and trust significantly affect GTSC optimization, especially as preferences increase. Additionally, a cost-sharing smart contract mechanism is designed to mitigate environmental investment's negative impact and optimize social welfare and product greenness.

Common methods used for topic modeling have generally suffered problems of overfitting, leading to diminished predictive performance, as well as a weakness towards reconstructing sparse topic structures that involve only a few critical words to aid in interpretation. Considering the text typically contained in customer feedback, this paper proposes a semiparametric topic model utilizing a two-step approach: (1) makes use of nonnegative matrix factorization to recover topic distributions based on word co-occurrences and; (2) use semiparametric regression to identify factors driving the expression of particular topics in the documents given additional auxiliary information such as location, time of writing, and other features of the author. This approach provides a generative model that can be useful for predicting topics in new documents based on these auxiliary variables, and is demonstrated to accurately identify topics even for documents limited in length or size of vocabulary. In an application to real customer feedback, the topics provided by our model are shown to be as interpretable and useful for downstream analysis tasks as with those produced by current legacy methods.

With digitisation globally on the rise, corporates are compelled to better understand the usage of their websites. In doing so, corporates will be empowered to better understand consumers, and make necessary adjustments to ultimately improve the corporate’s stance in the competitive global landscape of this modern age. However, the online website visit data has proven to be highly complex, big in data volume, and highly transactional with users expressing unique behaviours. Thus, extracting insight can be a complex problem to solve. This study aimed to employ unsupervised machine learning models to identify the intentions behind the visits on the observed website. The data studied was sourced from the Google Analytics tracking tool that was deployed on a corporate informative website. The study employed a k-means, hierarchical and dbscan unsupervised machine learning models to understand the intents behind visitors on the studied website. All three models detected five major intents that were expressed within the observed data. The intents identified were labelled as “accidentals”, “drop-offs”, “engrossed”, “get-in-touch” and “seekers”. On the observed data, all three unsupervised machine learning methods have performed well. However, in the context of the study, which investigated the intents that drove online visits, the hierarchical clustering method yielded superior results by maintaining the best balance between cluster homogeneity (stronger silhouette coefficients) and cluster size.