International Journal of Web Engineering and Technology (7 papers in press)

Regular Issues

• A Cultural Industry Text Classification Method Based on Knowledge Graph Information Constraints and Knowledge Fusion  
  by Xue Ji 
  Abstract: The study proposes a text classification method for the cultural industry. It uses knowledge graph information constraints and fusion. A knowledge graph is constructed for the cultural industry text, extracting entities and relationships with supervision. The encoding and decoding layers are optimised, and the knowledge fusion module incorporates attention. Article information is condensed and filtered, and the classifier calculates category probability. The experimental results show that in the loss function value test, in the validation data, our research method drops to the lowest value of 0.007 after about 25 iterations at the fastest, which is much lower than the lowest value of TextCNN about 0.038. When F1 value is tested in the validation data, when the average text length of our research method increases to 35 Byte, the highest F1 value reaches 0.88. Our research demonstrates effective text classification in the cultural industry with higher efficiency.
  Keywords: cultural industry; text classification; knowledge graph; attention mechanism; knowledge fusion.
  DOI: 10.1504/IJWET.2024.10062501
   
  
• DLSTMFRNN a newly developed network-based deep long short-term memory and recurrent neural network for stock market prediction  
  by Nagarjun Yadav Vanguri, Pazhanirajan S, Anil Kumar T. 
  Abstract: The stock market (SM) is fundamentally nonlinear in nature and the people invest in SM on the basis of predictions. The SM prediction is a highly challenging and complex process. The classical techniques may not guarantee the prediction reliability. Hence, a deep long short-term memory fused recurrent neural network (DLSTMFRNN) is presented for reliable SM prediction. Here, input time series data is obtained from the database and it is pre-processed utilising missing value imputation. Thereafter, features are extracted and then feds to the feature selection phase, in which the features are selected employing the Soergel metric. Finally, SM prediction is carried out utilising DLSTMFRNN, which is a new network designed by incorporating DLSTM and RNN. The DLSTMFRNN obtained a minimal mean absolute percentage error (MAPE) of 17.65, mean squared error (MSE) of 0.116, root mean square error (RMSE) of 0.341 and relative absolute error (RAE) of 0.156.
  Keywords: deep long short-term memory; DLSTM; recurrent neural network; RNN; stock market; SM; Soergel metric.
  DOI: 10.1504/IJWET.2024.10063128
   
  
• Application of improved K-means algorithm in the cultivation of creative music talents under the needs of sustainable development and transformation  
  by Peng Li, Zeng Fan 
  Abstract: In order to cultivate innovative personnel who are adapted to the development of university education, this paper proposes a K-means clustering algorithm (K-means) based on noise reduction autoencoder for the cultivation of creative music talents and explores the difficulties in cultivating innovative talents. The results show that the research-designed method outperforms K-means on the performance metrics NMI, AMI and FMI for the same dataset. The results of the practical application analysis show that the training of practical operation is weakened in talent training, and the emphasis on practical courses should be strengthened in the subsequent talent training plan.
  Keywords: sustainable development; autoencoder; K-means; talent training.
  DOI: 10.1504/IJWET.2024.10063583
   
  
• Deep learning-based task scheduling in edge computing  
  by Bantupalli Nagalakshmi, Sumathy Subramanian 
  Abstract: A potential paradigm called edge computing (EC) has recently come to light that supports internet of things (IoT) applications that are resource allocation with low latency services at the network edge. For scheduling the application tasks, the edge server's constrained processing capabilities present significant difficulties. The IoT-EC scenario is used in this research to study the task scheduling problem, and various jobs are scheduled to virtual machines (VMs) set up the edge server by maximising long-term task satisfaction. The proposed optimal task scheduling considers parameters like makespan, execution time, execution cost, and risk probability. Particularly, the risk probability estimation is done by the deep convolutional neural network (D-CNN). This estimation is based on task security and VM security. The scheduling of tasks is carried out via the new hybrid bald eagle Archimedes optimisation (HBEAO) by considering a multi-objective to minimise the makespan, execution time, execution cost, and risk probability. The proposed model is validated with existing models in terms of execution cost, execution time, fitness, makespan, risk probability, etc. It is observed that the HBEAO model attains less execution cost ($37.27), execution time (0.99 seconds), fitness (3.48%), risk probability (0.19%) and computation time (2,325.87 sec) respectively.
  Keywords: task scheduling; deep learning; edge computing; server; optimisation; internet of things; IoT; deep convolutional neural network; D-CNN.
  DOI: 10.1504/IJWET.2024.10061935
   
  
• PR-MQTT: a novel approach for traffic reduction and message prioritisation in IoT applications  
  by Jiby J. Puthiyidam, Shelbi Joseph 
  Abstract: IoT applications often involve devices with limited processing power, memory capacity, and low resource consumption. The vast number of devices connected to IoT networks generates massive amounts of data, making effective data management crucial for IoT applications. IoT applications prefer lightweight messaging protocols over the standard internet protocol, hyper text transfer protocol (HTTP). Message queue telemetry transport (MQTT) has emerged as a popular communication protocol for IoT applications. In some instances, specific messages may hold greater importance than others. However, most standard IoT protocols lack inherent mechanisms to prioritise incoming messages. This paper presents a new approach to reducing network traffic in IoT applications by selectively transmitting messages while prioritising the processing of urgent messages. The proposed method is integrated with HBMQTT, an MQTT broker. The experimental evaluation indicates that with the proposed PR-MQTT broker, the latency of the priority messages remains nearly constant, irrespective of the message's index position. Priority messages are consistently delivered within 10-15 milliseconds, resulting in a speed improvement of over 90% compared to regular messages. Additionally, the proposed approach reduces CPU resource utilisation and network traffic by 25% and the transmission delay of normal messages by 50%.
  Keywords: internet of things; IoT; MQTT protocol; message priority; IoT networks; network traffic.
  DOI: 10.1504/IJWET.2024.10061282
   
  
• A distributed framework for distributed denial-of-service attack detection in internet of things environments using deep learning  
  by Wawire Amisi Silas, Lawrence Nderu, Dalton Ndirangu 
  Abstract: Internet of things (IoT) networks dominate industries, homes, organisations, and other aspects of life owing to their automation capabilities. However, IoT networks are vulnerable to attacks, especially distributed denial-of-service (DDoS) attacks, as they tend to have low computational capabilities and are highly diverse. While current research shows the potential of utilising deep learning methods to detect DDoS attacks, there is a lack of a framework that can be used to deploy an effective deep learning algorithm to detect DDoS attacks in heterogeneous IoT environments. Accordingly, this paper developed a DDoS detection framework based on the CNN-BiLSTM model, which can be deployed in a distributed network and includes adequate pre-processing. Simulations were also done to demonstrate the application of the framework and its effectiveness.
  Keywords: machine learning; artificial intelligence; internet of things; IoT; deep learning; convolutional neural networks; CNNs; BiLSTM; distributed denial-of-service; DDoS.
  DOI: 10.1504/IJWET.2024.10062503
   
  
• Performance evaluation of higher education management under the background of knowledge management  
  by Xun Mo 
  Abstract: In view of the shortcomings of the accuracy and objectivity of the current higher education management performance evaluation methods under the background of KM, this paper studied and constructed the education management performance evaluation model. On this basis, a back propagation neural network (BPNN) model based on the improved whale optimisation algorithm (IWOA) was proposed for fitting the index data. The experimental results showed that the number of iterations required by the IWOA-BPNN model was only 68; the F1 value was 0.961; the recall value was 0.950; the fitness degree was 0.948; the MSE was 0.463; the MAE was 8.53; the accuracy rate was 0.985 and the AUC value was 0.912, all of which were superior to the most advanced intelligent evaluation method of educational management performance. The above results show that the evaluation model based on IWAO-BPNN can accurately and effectively realise the intelligent evaluation of educational management performance.
  Keywords: knowledge management; KM; performance evaluation; indicator system; back propagation neural network; BPNN; improved whale optimisation algorithm; IWOA.
  DOI: 10.1504/IJWET.2024.10062982