Intelligent Systems with Applications最新文献

In the process of building domain knowledge graph, the result of relationship extraction between entities is an important guarantee of the quality of the graph. Therefore, we propose a clustering method based on reinforcement learning for remote supervised relation extraction. For the relationship extraction of accident information in the aviation domain mapping, a clustering method combining local dense and global dissimilarity is proposed in combination with remote supervision, which can obtain a large amount of low-noise labeled data and reduce part of the wrong labeling and missing labeling due to the strong specialization in the aviation domain; meanwhile, reinforcement learning is introduced to denoise the negative instance noise in the positive sample data; Finally, we propose a two-attention segmentation (DAPCNN) relationship extraction model to mine deep semantic sentences. The experimental results show that in the civil aviation relationship extraction text constructed in this paper, the Micro_R, Micro_P and Micro_F1 values of the proposed relationship extraction method reach 83.41 %, 84.16 % and 83.96 %. In the open relationship extraction dataset DuIE, The Micro_R, Micro_P and Micro_F1 of the proposed method are up to 83.41 %, 93.58 % and 94.02 % respectively. Compared with the current advanced multi-instance and multi-label model, the proposed method can more accurately extract the relationship between aviation accident entities. At the same time, the performance of the open data set is also good, and has a certain universality.

With the swift advancement of online teaching in vocational education, an increasing number of web-based course materials are being made available to students, granting them the freedom to select resources that suit their personal needs. To optimize the effectiveness of artificial intelligence-enabled smart vocational education, this study presents a course recommendation model centered on learning behaviors and interests. The model utilizes short-term preferences reconstruction behavior contribution to identify fluctuations in learners' interests in real-time. A model for recommending courses is proposed based on short-term preferences and enhancements to learning behavior. Its purpose is to tackle the issue of generalization arising from sparsity and weak correlation in learning behavior. The outcomes demonstrated the model put forth in the study achieved higher Hit Rate (HR) and Normalized Discounted Cumulative Gain (NDCG) values in comparison experiments with multiple models. Hence, this suggested that creating a novel component of historical learning behavior, powered by dynamic interest factors, could resolve the issue of changing learning interests and enhance the efficacy of course recommendation models. Furthermore, the introduction of a correlation mapping network enables the forward mapping transformation from weak to strong learning behavior, thus improving and optimizing input for the agent strategy, reducing data sparsity, and enhancing the performance and generalization of the course recommendation model.

Curriculum Learning (CL) is an effective technique to train machine learning models where the training samples are supplied to the model in an easy-to-hard manner. Similar to human learning, the model can benefit if the data is given in a relevant order. Based on this notion, we propose to apply the concept of CL to the task of session-based recommender systems. Recurrent Neural Networks and transformer-based models have been successfully utilized for this task and shown to be very effective. In these approaches, all training examples are supplied to the model in every iteration and treated equally. However, the difficulty of a training example can vary greatly and the recommendation model can learn better if the data is given according to an easy-to-difficult curriculum. We design various curriculum strategies and show that applying the proposed CL techniques to a given recommendation model helps to improve performance.

With the continuous development of network technology, online education has become an important form of education. However, in the online education model, it is difficult for educators to effectively evaluate students' learning status, and using a learning status evaluation model can effectively solve this problem. The main goal of this model is to comprehensively evaluate students' learning behavior, progress, and outcomes, in order to understand their learning status, provide effective teaching feedback to teachers, help students improve learning methods, and improve learning efficiency. The current automatic evaluation model for student learning status has certain limitations in terms of applicability and accuracy. A student learning state evaluation model based on Multi task Cascaded Convolutional Networks (MTCNN) is proposed to address the effectiveness of online education student learning state evaluation. Use the facial image acquisition function to extract students' facial features, process each feature through label classification, and then analyze the students' attention and learning emotions. Finally, analyze the effectiveness of the research method application. The results showed that the train_loss value of the learning state evaluation model proposed in the study can be reduced to about 0.1; the train_acc value can reach more than 95 %, and the overall volatility is small; the overall evaluation accuracy of facial expressions can reach 74.71 %, which is significantly better than cpc, VGG19 and other evaluation methods; compared with the comprehensive evaluation results and multi-modal analysis methods, only two evaluations at the critical value are different. The experimental results show that the online education students’ learning status evaluation model designed by the research has a high accuracy rate and has a certain application potential in the field of online education.

With the rapid popularization and application of Internet of Things technology, smart devices have become an indispensable part of people's daily lives. Therefore, it is crucial to accurately identify these devices as their numbers continue to grow. The research aimed to introduce a lightweight method for identifying Internet of Things devices based on network flow characteristics and scheduling algorithms. This can improve device identification accuracy while maintaining high efficiency. The research constructed a comprehensive optimization algorithm selection framework to achieve performance optimization in different scenarios. This framework took into account many factors, including network traffic characteristics, device identification requirements, and system efficiency, to ensure flexible adaptation in different scenarios and optimize overall performance. Research results showed that the proposed system had an accuracy of 96.8 % at 1-minute intervals, which increased to 99.7 % at 10-minute intervals, and reached 99.9 % at both 30-minute and 60-minute intervals. In 100 experiments, the research method improved the accuracy by an average of 1.5 % compared with the baseline. In fingerprint recognition, the overall accuracy of the long short-term memory network exceeded 90 %, with an area under the curve of 0.99. Most devices had an accuracy of over 95 % in recognition, and the recall rate remained around 90 %, the effectiveness of the method proposed in the study was further verified. The method proposed in the study not only improved the accuracy and efficiency of device recognition, but also provided powerful solutions for the field of network security. This provides useful guidance for research and practical applications in related fields.

Ultrasound imaging has become one of the most frequently employed modalities to detect and classify breast irregularities, which is a relatively cost-effective and important complement to mammography. To assist radiologists in locating worrisome lesions and improving the accuracy of diagnosis, a computer-aided diagnosis system is proposed which incorporates the knowledge of Generative Adversarial Network (GAN), weighted average based ensemble technique, and feature fusion based ensemble technique. After performing encoder decoder based lesion segmentation incorporating weighted ensemble architecture, a dual-staged feature fusion-based stacked ensemble meta-classifier architecture is employed for the final classification where three deep neural network branches are employed, and the generated feature maps from those branches were fused and fed to the fully connected network to achieve the final diagnosis result. The residual learning architecture and the pretrained foundation made the system faster, whereas compound scaling and ensemble architecture boosted the overall performance. The proposed methodology is evaluated on the BUSI, UDIAT, and Thammasat datasets. The Dice score reached to 0.8397, and the IoU score reached to 0.7482 in segmentation on the benign lesions of BUSI dataset whereas the classifier achieved a highest accuracy of 99.7%, F1-score of 99.7%, and AUC score of 0.999 in classification on the BUSI dataset. The results on the UDIAT and Thammasat datasets also indicate that our proposed method shows better performance than other methods. Thus, the proposed architecture can be considered for easy and automated diagnosis purposes.

BIM tools enable decision-making during the lifecycle of engineering structures, such as bridges, tunnels, and roads. National Road Authorities use Asset Management Systems (AMS) to manage and monitor operational information of assets from European Highways, including access to sensor and inspection data. Interoperability between BIM and AMS systems is vital for a timely and effective decision-making process during the operational phase of these assets. The European project Connected Data for Effective Collaboration (CoDEC) designed a framework to support the connections between AMS and BIM platforms, using linked data principles. The CoDEC Data Dictionary was developed to provide standard data formats for AMS used by European NRA. This paper presents the design and development of an Engineering Structures ontology used to encode the shared conceptualization provided by the CoDEC Data Dictionary. The ontology is evaluated, validated, and demonstrated as a base for data exchange between BIM and AMS.

Tuberculosis (TB) remains a global health threat, and rapid, automated and accurate diagnosis is crucial for effective control. The tedious and subjective nature of Ziehl-Neelsen (ZN) stained smear microscopy for identifying Mycobacterium tuberculosis (MTB) motivates the exploration of alternative approaches. In recent years, machine learning (ML) methods have emerged as promising tools for automated TB detection in ZN-stained images. This systematic literature review (SLR) comprehensively examines the application of ML methods for TB detection between 2017 and 2023, focusing on their performance metrics and employed dataset characteristics. The study identifies advancements, establishes the state of the art, and pinpoints areas for future research and development in this domain. It sheds light on the discussion about the readiness of machine-learning methods to be confidently, reliably and cost-effectively used to automate the process of tuberculosis detection in ZN slides, being it significant for the health systems worldwide.

Following established SLR guidelines, we defined research questions, retrieved 175 papers from 7 well-known sources, and discarded those not complying with the inclusion criteria. Data extraction and analysis were performed on the resulting 65 papers to address our research questions. The key contributions of this review are as follows. First, it presents a characterization of the state of the art of ML methods for ZN-stained TB detection, especially in sputum and tissue. Second, it analyzes top-performing methods and pre-processing techniques. Finally, it pinpoints key research gaps and opportunities.

The study proposes a cascaded hybrid neural network commodity demand prediction model based on multimodal data. This model aims to improve the accuracy of commodity demand forecasts on e-commerce platforms. By constructing multimodal data feature clusters and utilizing a spatial feature fusion strategy, historical order information, and product evaluation sentiment data are integrated. The model combines the advantages of bi-directional long and short-term memory networks and bi-directional gated recurrent unit networks. The proposed cascaded hybrid strategy-based model significantly enhances accuracy in commodity demand forecasting. Results indicated an average absolute error of 0.1682 and root mean square error of 0.4537 for weekly commodity forecasts. For long-term commodity demand, the average absolute error was 0.8611 with a root mean square error of 8.1938. These outcomes highlight the algorithm's high prediction accuracy, making it valuable for commodity demand prediction on e-commerce platforms and providing a framework for effective inventory management.

Accurate real-time prediction of link quality is crucial for enhancing the reliable responsiveness of wearable devices within Wireless Wearable Sensor Networks (WWSNs). Specifically, the Signal-to-Noise Ratio (SNR), a pivotal parameter for predicting link quality, exhibits complex temporal characteristics influenced by stochastic and non-stochastic factors. To improve the accuracy of link quality prediction in WWSNs, we aim to explore a novel predictive model, introducing a filtering layer that seeks to enhance the precision of predicting upper and lower boundaries of link reliability confidence intervals. First, we adopt the SNR time series as the evaluation metric and decompose the SNR sequences into time-varying and stochastic standard deviation sequences by wavelet decomposition. Subsequently, we propose an innovative SCNN-LSTM model, incorporating the SincNet filtering layer to extract specific frequency components from the input SNR sequences. Afterward, integrating standard deviation sequences, the model predicts upper and lower boundaries of link reliability confidence intervals. Finally, we conduct the validation experiments on the public dataset LightGBM-LQP and our WWSN dataset Basketball shot. Compared to BPNN, ARIMA, and WNN, the evaluation matrices of MAE, RMSE, R2 in SCNN-LSTM have been improved, and the deviation between the predicted standard deviation and the actual standard deviation has reached the minimum of 0.1. The results demonstrate that SCNN-LSTM outperforms classical prediction models in predicting upper and lower limits of link reliability confidence intervals in WWSNs.