International Journal of Grid and Utility Computing (6 papers in press)

Regular Issues

• Recommendation system based on space-time user similarity
  by Wei Luo, Zhihao Peng, Ansheng Deng 
  Abstract: With the advent of 5G, the way people get information and the means of information transmission have become more and more important. As the main platform of information transmission, social media not only brings convenience to people's lives, but also generates huge amounts of redundant information because of the speed of information updating. In order to meet the personalised needs of users and enable users to find interesting information in a large volume of data, recommendation systems emerged as the times require. Recommendation systems, as an important tool to help users to filter internet information, play an extremely important role in both academia and industry. The traditional recommendation system assumes that all users are independent. In this paper, in order to improve the prediction accuracy, a recommendation system based on space-time user similarity is proposed. The experimental results on Sina Weibo dataset show that, compared with the traditional collaborative filtering recommendation system based on user similarity, the proposed method has better performance in precision, recall and F-measure evaluation value.
  Keywords: time-based user similarity; space-based user similarity; recommendation system; user preference; collaborative filtering.
  
  
• Joint end-to-end recognition deep network and data augmentation for industrial mould number recognition  
  by RuiMing Li, ChaoJun Dong, JiaCong Chen, YiKui Zhai 
  Abstract: With the booming manufacturing industry, the significance of mould management is increasing. At present, manual management is gradually eliminated owing to need for a large amount of labour, while the effect of a radiofrequency identification (RFID) system is not ideal, which is limited by the characteristics of the metal, such as rust and erosion. Fortunately, the rise of convolutional neural networks (CNNs) brings down to the solution of mould management from the perspective of images that management by identifying the digital number on the mould. Yet there is no trace of a public database for mould recognition, and there is no special recognition method in this field. To address this problem, this paper first presents a novel data set aiming to support the CNN training. The images in the database are collected in the real scene and finely manually labelled, which can train an effective recognition model and generalise to the actual scenario. Besides, we combined the mainstream text spotter and the data augmentation specifically designed for the real world, and found that it has a considerable effect on mould recognition.
  Keywords: mould recognition database; text spotter; mould recognition; data augmentation.
  
  
• Dynamic parameter optimisation method for database based on large language model and evolutionary reinforcement learning  
  by Lihua Pan, Jin Li 
  Abstract: This study proposes an intelligent tuning framework that integrates large language models with evolutionary reinforcement learning for dynamic database parameter optimisation. By leveraging the semantic understanding of LLMs for initialisation, exploration guidance, and adaptive reward weighting alongside evolutionary strategies, the framework enables efficient online adaptation in high-dimensional parameter spaces. Validated across OLTP, HTAP, and cloud-native scenarios, the method improves throughput by 96.9%, reduces latency by 54.1%, and enhances resource utilisation by 28.7% compared to conventional reinforcement learning, while also accelerating convergence and reducing total tuning time. Ablation studies confirm the critical contribution of LLM-driven collaborative mechanisms to overall performance gains.
  Keywords: large language model; evolutionary reinforcement learning; database parameter tuning; dynamic optimisation; intelligent decision-making; performance optimisation.
  DOI: 10.1504/IJGUC.2026.10078071
   
  
• GLE: an important patentee identification method based on comprehensive structural entropy  
  by Na Deng, Jiu-an Zhang 
  Abstract: Promoting patent cooperation among universities, enterprises, and research institutes transforms academic knowledge into scientific achievements, driving industrial innovation. Identifying important nodes (patentees) in patent cooperation networks allows public resources to support core patentees, improving Industry-University-Research (IUR) cooperation efficiency. We reframe this as a node importance identification problem within patent innovation networks. Since existing algorithms rarely consider both global and local network topology simultaneously, we propose an improved structural entropy algorithm to identify core patentees. The method computes weighted centrality based on node degree and strength, calculates local structural entropy, and incorporates global position information via the K-shell method. Node importance is then determined by combining local structural entropy with neighbouring node contributions. Using a patent cooperation network of Hubei universities, the SIR propagation model and Kendall correlation coefficient validate our approach. Results confirm the method evaluates node importance more effectively and accurately than existing algorithms.
  Keywords: patent collaboration network; node importance; structural entropy; weighted complex network; SIR.
  DOI: 10.1504/IJGUC.2024.10078257
   
  
• Distributed network security authentication mechanism integrating blockchain and artificial intelligence  
  by Long Li, Jinka Wang, Junli Luo 
  Abstract: This paper proposes BlockGrad, a blockchain-based deep learning framework that enhances federated learning security by integrating Byzantine Fault Tolerance (BFT) consensus and reputation-based weighted aggregation. Unlike traditional methods such as Federated Averaging and Multi-Krum, BlockGrad combines gradient validation, reputation evaluation, and dynamic node replacement into a closed-loop defence mechanism. Through clustering-based gradient screening and weighted aggregation, the framework effectively mitigates the impact of malicious participants. Simulation results under a 30% malicious-node attack demonstrate that BlockGrad achieves higher and more stable classification accuracy than Federated Averaging, showing strong robustness in adversarial environments. The proposed approach also enables decentralised trust management and long-term behavioural tracking, which are absent in many existing robust aggregation methods. Furthermore, the computational overhead of BlockGrad is analysed, and its applicability to large-scale distributed systems is discussed. Future work will focus on improving scalability, privacy protection, and algorithmic efficiency.
  Keywords: blockchain; artificial intelligence; deep learning; Byzantine fault tolerance; distributed network security.
  DOI: 10.1504/IJGUC.2026.10079272
   
  

Special Issue on: Cloud and Fog Computing for Corporate Entrepreneurship in the Digital Era

• Study on the economic consequences of enterprise financial sharing model  
  by Yu Yang, Zecheng Yin 
  Abstract: Using enterprise system ideas to examine the business process requirements of firms, the Financial Enterprise Model (FEM) is a demanding program. This major integrates finance, accounting, and other critical business processes. Conventional financial face difficulties due to low economic inclusion, restricted access to capital, lack of data, poor R&D expenditures, underdeveloped distribution channels, and so on. This paper mentions making, consuming, and redistributing goods through collaborative platform networks. These three instances highlight how ICTs (Information and Communication Technologies) can be exploited as a new source of company innovation. The sharing economy model can help social companies solve their market problems since social value can be embedded into their sharing economy cycles. As part of the ICT-based sharing economy, new business models for social entrepreneurship can be developed by employing creative and proactive platforms. Unlike most public organizations, double-bottom-line organizations can create social and economic advantages. There are implications for developing and propagating societal values based on these findings.
  Keywords: finance; economy; enterprise; ICT; social advantage.