Neural relationship extraction is an important task in natural language processing, aimed at extracting relationships between target entity pairs from a given text. In recent years, with the development of deep neural networks, various types of neural networks to extract sentence entity-level, fragment-level, and sentence-level features for relationship extraction have become a mainstream research direction. Most existing studies use the BERT model to embed sentences and then use CNN to manipulate all words in the entire sentence to obtain fragment-level features. This article proposes a new word-centered context fragment-level method based on pruning the shortest dependency path between entity pairs. We demonstrate that using a pruning method based on the shortest dependency path between entity pairs can effectively improve the ability of model fragments and information extraction. We evaluated our method on a public benchmark: SemEval 2010 Task 8. The experimental results show that our method outperforms the advanced model using BERT as the embedding.
This study introduces a framework for clustering competition coevolution optimization algorithm based on the parallel Lion Swarm Optimization Algorithm (LSO). This framework combines clustering and competitive coevolution concepts under existing parallel computing paradigms. Initially, clustering categorizes particles of the total population, followed by parallel computing principles where particles within each classified subpopulation undergo local optimization using distinct optimization mechanisms. After a certain number of iterations, these subpopulations coevolve through an island-based topology. Experimental results demonstrate significant advantages of the proposed algorithm over traditional methods in both CEC2013 benchmark functions and feature selection problems, affirming its potential and effectiveness in practical applications. This framework introduces a novel approach and method for addressing complex problems, offering broad prospects for application.
A group decision making approach considering the correlation of multiple evaluation attributes under information described by Z-numbers is proposed. The approach primarily consists of two parts: (1) obtaining the group preference matrix under information described by Z-numbers; (2) addressing the interactions among multiple evaluation attributes. To achieve this, firstly, a technique for ranking Z-numbers is proposed. The provided ranking technique better distinguishes Z-numbers. Additionally, the Choquet integral for Z-numbers (ZN − CIβ) is introduced. The ZN − CIβ captures the interconnections among multiple evaluation attributes in an uncertain environment. Secondly, the weights of decision makers are determined through Shapley values, ensuring fairness in the decision making process. In addition, the comprehensive group decision matrix is formed by aggregating the matrices of individual decision makers, each represented by their respective Z-numbers. Finally, the efficacy of the approach presented in this paper is validated through a case of selecting B&Bs.
Determining Hopf bifurcation points is an important task in the study of nonlinear dynamic systems, but existing methods require a large amount of computation and have low ratios of success. In this study, we present a numerical method to determine Hopf bifurcation points: the Split Iteration Technique and prove its convergence. Compared with existing methods, this method results in much less computation and a much higher ratio of success.
The study proposes a novel approach that uses block pulse functions (BPFs) with constant delays to approximate stochastic Volterra integral equations (SVIEs). The method simplifies the problem by converting delay-containing SVIEs into algebraic ones using operational matrices of BPFs. This approach is easily solvable and effective, as shown through numerical examples.
Andrographis paniculata extract (APE), containing andrographolide, has long been used in Thai traditional medicine for its antiviral properties, including against SARS-CoV-2. The recommended dose in Thailand is 180 milligrams three times daily. We developed a mathematical model incorporating the pharmacokinetic/pharmacodynamic (PK/PD) profile of andrographolide therapy to predict viral dynamics in four COVID-19 patients. APE was found to enhance the efficacy of COVID-19 treatment. Despite similar infection behavior, there were differences in the quality of the virus peak between treated and untreated patients.
As outbreaks of respiratory illnesses continue to arise, such as influenza and the JN.1 variant of COVID-19, this study utilizes YOLOv7 in an effort to investigate the effectiveness of face mask detection in a crowded setting. In considering the practicality of implementing this in real time, problems may arise in ensuring consistency in detecting instances in various settings with environmental factors such as lighting conditions, crowd density and individual's activity. With the help of transfer learning and a dataset procured to simulate a crowded setting, this study thoroughly assesses YOLOv7 model's performance to distinguish its advantages and disadvantages. The results show that the model performs better at lower confidence levels, but they also point to a bias in the identification of unmasked people, which may be related to problems with data imbalance. The study makes recommendations centered on dataset balancing and hyperparameter adjustment to solve these drawbacks and improve real-world applicability. This study highlights the critical role of technological innovation and advancements in addressing contemporary health crises and advancing public health interventions, in addition to contributing to ongoing efforts to combat respiratory illnesses. This is achieved through the research's bridging of theoretical advancements with practical insights.
The paper focuses on investigating stock selection strategies based on decision tree and multi-factor models. The study explores the utilization of decision trees to incorporate multiple factors in the stock selection process, aiming to enhance the accuracy and effectiveness of investment decisions. The methodology involves collecting a comprehensive dataset of financial variables and factors that are known to influence stock returns. A decision tree algorithm is applied to construct a predictive model by recursively partitioning the dataset based on selected factors. The resulting decision tree provides a systematic framework for identifying key factors and making informed stock selection choices. Comparative analysis is performed against benchmark indices and traditional stock selection approaches to gauge its added value and potential for generating superior returns. Based on the relevant data of the constituent stocks of the CSI300 Index, significant and effective factors are selected to construct different classification decision tree models. Empirical evidence suggests that machine learning algorithms can effectively predict stock returns. Backtesting is conducted using stock return data from 2022 to 2023, and it is found that the decision tree and multi-factor stock selection model achieved the objective of generating excess returns compared to the CSI300 Index.
Little is known about the association of social media and belief in alcohol and cancer with binge drinking. This study aimed to perform feature selection and develop machine learning (ML) tools to predict occurrence of binge drinking among adults in the United State. A total of 5,886 adults including 1,252 who ever experienced with binge drinking were selected from the 2022 Health Information National Trends Survey (HINTS 6). Feature selection of 69 variables was conducted using Boruta and the Least Absolute Shrinkage and Selection Operator (LASSO). The Random Over Sampling Example (ROSE) method was utilized to deal with the imbalance data. Seven machine learning (ML) tools including the Support Vector Machines (SVMs) algorithms, Logistic Regression, Naïve Bayes, Random Forest, K-Nearest Neighbor, Gradient Boosting Machine, and XGBoost were applied to develop ML models to predict binge drinking. The overall prevalence of binge drinking among U.S. adults is 21.3%. Both Boruta and LASSO selected 28 identical variables. SVM with Radial Basis Function revealed the best model with the highest accuracy of 0.949 and sensitivity of 0.958. The top risk factors of binge drinking were tobacco use (e-cigarette use and smoking status), belief in alcohol (alcohol decreases the risk of future health), belief in cancer (prevention is not possible, worry about getting cancer), and social media (social media visits and sharing health information). These findings underscore the need for multiple health behavior interventions to enhance education related to alcohol use and cancer and how to effectively employ social media to improve health outcomes.
Sequent calculi for applied logics have nonstandard logical inference rules and various axioms, cut is essential in them. The following constraints on derivations in these calculi do not compromise completeness. Contraction rules follow cut and logical rules. Weakening rules precede logical rules. Consecutive cut rules are ordered so that larger formulas are cut first. Consecutive standard logical rules are ordered so that their principal formulas are in an increasing order.
Agricultural development level is an important link of economic and social development. Guangdong Province is a large agricultural province in China, this paper takes the assessment of the agricultural development level of cities in Guangdong Province as the research object, which helps to understand the differences in the level of agricultural development in different places and puts forward corresponding policy suggestions, and has important research significance and application value. This paper uses the hierarchical cluster analysis and principal component analysis to study the agricultural development level in each prefecture-level city in Guangdong Province. First of all, the prefecture-level cities in Guangdong Province are divided into four categories by hierarchical cluster analysis. Then, four principal components are obtained by principal component analysis, and the prefecture-level cities in Guangdong Province are ranked. Combining the results of hierarchical cluster analysis and principal component analysis, the agricultural development level in Guangdong Province can be divided into three categories. The results show that, overall, most regions in western and northern Guangdong are ranked at the top, most regions in the Pearl River Delta are ranked in the middle, and eastern Guangdong is ranked at the bottom. Among them, under the circumstances of insufficient allocation of agricultural water resources, western Guangdong has achieved a high level of agricultural development by developing intensive agriculture and ecological agriculture, and so on. In addition, this paper puts forward corresponding policy suggestions around the above three types of regions.
Though the interest in learning IoT and AI involved in industry 4.0 is dramatically increasing in recent years, many universities have not integrated those up-to-date technologies and applications in engineering education. In this research, many aspects of IoT and AI applied in industry 4.0 are introduced in engineering curriculum development. The redesigned curriculum with innovative technology is developed for students to learn IoT, AI, machine learning, and smart sensors; and prepare engineering students in the demanding field of IoT, AI, and Industry 4.0 using new technologies, coupled with industry-based and problem-based hands-on learning. This work could have great impact on activating student interest and enthusiasm for learning engineering in comprehensive application and creativity.
In this paper we present COVIDTran, an automated COVID diagnostic system that takes symptomatic cough audios as input and identifies potential cases of COVID19. Adopting principles from Transfer Learning, we implement neural network based on Vision Transformer that processes the spectrographic maps of the cough audio signals, and promote the robustness of our model by integrating contextual information from similar flu symptomatic datasets via transfer learning. Experimental results involving crowdsourced COVID coughing and speech datasets suggest that our strategy outperforms other current methods as measured by different metrics, thereby providing new insights on automated COVID19 diagnosis on top of existing methods.
Vector graphics has been employed in a wide variety of applications due to its scalability and editability. A drawing in a standard vector graphics file, such as SVG (Scalable Vector Graphics), is composed of numerous paths, and a path typically consists of a list of Bézier curves. Traditional vector graphics watermarking techniques, such as Fourier and wavelets, require the initial step of sampling the Bézier curve to obtain a large number of discrete points. However, these methods allow only one line to be selected for watermarking. In this paper, we introduce a novel watermarking of 2D vector graphics based on a new class of orthogonal function system(V-system). It can exactly represent all curves in a vector graphics by a global format directly, eliminating the need for sampled processing. Furthermore, multiple curves can be easily watermarked simultaneously by modifying the V descriptors. A watermark embedded using this method can be successfully extracted even under transformation attacks. Experimental results confirm the imperceptibility and robustness of the proposed method.
Ensemble learning is one of the most studied topics in classification domain, it is proven that ensemble learning is effective for classification tasks with multiple labels. Nevertheless, achieving accurate predictions for data with varying dimensions and characteristics remains a formidable task. Enhancing the generalization capabilities of ensemble classifiers poses a significant challenge in the field of ensemble learning. To address this issue and improve accuracy, we propose a novel ensemble method. In our method, a combination of deep learning classifier (i.e., CNN, Bi-LSTM) and classical classifiers (e.g., KNN, SVM, Naive Bayes, Dtree) is constructed and then the optimal weighting parameters for the base classifier are assigned. We enhance the model's robustness by employing lasso regularization to reduce data dimensions. The results on 10 folds show that our method can significantly enhance the model's generalization performance and precision in classification tasks. Our model consistently outperforms traditional and existing ensemble methods across 18 publicly available datasets, as evidenced by accuracy and the Friedman test.
Investigating the application of deep learning methodologies, specifically focusing on the YOLOv8-based vehicle type detection, to extract pertinent information accurately from dashcam footage. YOLOv8 stands out for its swift and precise object recognition capabilities, serving as a foundational element for constructing a robust system capable of identifying various vehicle types across diverse environmental conditions and scenarios. The dataset, derived from dashcam footage, encompasses varying types and amounts of vehicles captured in the evening and highway setting. Utilizing Roboflow, the dataset is prepared for training and testing, with vehicle types pre-determined for each frame. The YOLOv8 algorithm, renowned for its one-stage detection paradigm, is enhanced with advancements in accuracy and speed. Key components of YOLOv8, including the backbone network, Feature Pyramid Network (FPN), and anchor boxes, contribute to its proficiency in object detection. The YOLOv8 model demonstrates high accuracy rates for identifying vehicle types, with individual accuracies exceeding 90%. This study underscores the efficacy of YOLOv8 in vehicle-type detection and its potential for enhancing surveillance and traffic management systems.
This project proposes an operational predictive solution that assists in predicting the flood occurrence at Kota Belud in Sabah, particularly in estimating and predicting the water level, and affected area. It is motivated by the work under the Security And Integrated Flood Operation Network at Kota Belud (S.A.I.F.O.N@Belud). To anticipate the likelihood of a flood, this solution uses machine learning. The suggested method needs input from the S.A.I.F.O.N@Belud system's current sensors. The created solution can assist in determining forecast lead times that enable the authorities to issue prior notice in order to tackle the impending flood catastrophe. The algorithm consists of a Long Short-Term Memory (LSTM) neural network model that can predict the water level of the Tempasuk River in Sabah for the next 30 minutes. The predicted value is then fed into a rule-based flood risk model to predict the flood risk status in the next 30 minutes. The LSTM prediction model developed using data from S.A.I.F.O.N@Belud dataset yields an average of root mean squared error and the mean absolute error of 0.08 and 0.03 respectively. Whereas, the rule-based flood risk model achieves an overall accuracy of 98.18%. The proposed model lays the foundation in expanding the S.A.I.F.O.N@Belud to prediction phase beyond the real time monitoring.
Mitral Transcatheter Edge-to-Edge Repair (mTEER) is a medical procedure utilized for the treatment of mitral valve disorders. However, predicting the outcome of the procedure poses a significant challenge. This paper makes the first attempt to harness classical machine learning (ML) and deep learning (DL) techniques for predicting mitral valve mTEER surgery outcomes. To achieve this, we compiled a dataset from 467 patients, encompassing labeled echocardiogram videos and patient reports containing Transesophageal Echocardiography (TEE) measurements detailing Mitral Valve Repair (MVR) treatment outcomes. Leveraging this dataset, we conducted a benchmark evaluation of six ML algorithms and two DL models. The results underscore the potential of ML and DL in predicting mTEER surgery outcomes, providing insight for future investigation and advancements in this domain.
Uplift modeling has been used effectively in fields such as marketing and customer retention, to target those customers who are more likely to respond due to the campaign or treatment. Essentially, it is a machine learning technique that predicts the gain from performing some action with respect to not taking it. A popular class of uplift models is the transformation approach that redefines the target variable with the original treatment indicator. These transformation approaches only need to train and predict the difference in outcomes directly. The main drawback of these approaches is that in general it does not use the information in the treatment indicator beyond the construction of the transformed outcome and usually is not efficient. In this paper, we design a novel transformed outcome for the case of the binary target variable and unlock the full value of the samples with zero outcome. From a practical perspective, our new approach is flexible and easy to use. Experimental results on synthetic and real-world datasets obviously show that our new approach outperforms the traditional one. At present, our new approach has already been applied to precision marketing in a China nation-wide financial holdings group.
In recent years, Deep CNN (DCNN) models have achieved great success in the field of computer vision. However, such models are still considered to lack interpretability. One of fundamental issues underlying this problem can be noted as follows: The decision-making of a DCNN model is considered as a “black-box” operation. In this study, we propose to use binary tree structure convolution layers (TSCL) to interpret the decision-making mechanism of a DCNN model in the image recognition task. First, we design a TSCL module, in which each parent layer generates two child layers, and then integrate them into a normal DCNN. Second, we design an information coding objective to guide each two child nodes of one parent node to learn the particular information coding that we expected. Through the experiments, we can verify that: 1) the logical process of decision-making made by ResNet models can be explained well based on the "decision information flow path" formed in the TSCL module; 2) the decision-path can reasonably interpret the decision reversal mechanism (Robustness mechanism) of the DCNN model; 3) the credibility of decision-making can be measured by the matching degree between the actual and expected decision-path.
