M. Rashmi

Author-level bibliometric research drift is a gradual change in research focus over a specific time period due to emerging technologies, paradigm shifts in science, or shifting societal needs. Author-level bibliometric research drift must be identified for the comprehension of scientific progress trends, consistency in primary research objectives, and optimization of resource utilization. This study suggests a systematic approach to identify and quantify the Author-level bibliometric research drift through the aid of scholarly articles, authors’ contributions, and evolving nature of keyword distributions. The proposed framework measures author-level bibliometric research drift rather than semantic evolution of scientific concepts. Using natural language processing and deep learning techniques, we track changes in the research topic and project them onto a latent semantic structure. We demonstrate that the proposed TADGLN-LSTM, outperforms traditional approaches in identifying Author-level bibliometric research drift. Dynamic GNNs, with their ability to capture evolving interactions between authors, topics, and publications, excel in capturing temporal relationships and structural changes in the research domain. The implications of Author-level bibliometric research drift are profound, impacting funding, collaboration networks, and innovations particular to specific disciplines. Our approach enables institutions and policy makers to quantify the trends of research domains, thereby ensuring perpetual relevance and strategic foresight, ultimately facilitating a more responsive and forward-looking research ecosystem.

By depicting items as nodes and their connections as links, networks or graphs attract more attention in complex systems as a way to simulate real-world interactions. Finding influential nodes in human contact networks or other social networks is essential to comprehending disease transmission, which is dependent on the frequency and intensity of contact. Nevertheless, the majority of current research ignores the variability of real-world interactions in favour of uniform connection strength. To address this gap, we propose a unified influence estimation model (uiem) that integrates multi-hop diffusion, local structural reinforcement, and interaction diversity into a single adaptive framework. The model constructs a weighted graph where edge weights reflect interaction frequency/ real weights based on the different scenarios. One-hop and two-hop components capture direct and indirect diffusion influence, while the local structural reinforcement index (lsri) quantifies a node’s connectivity strength and connections within its neighborhood. Additionally, a diversity-weighted fusion (dwf) mechanism combines weighted degree and local clustering entropy (lce) to balance structural intensity and interaction diversity. Experimental results on multiple human contact networks and social networks demonstrate that uiem outperforms existing methods, effectively identifying structurally and functionally influential nodes and providing deeper insights into the dynamics of real-world contact-based spreading processes.

Plant diseases are a serious issue that cause food shortages and financial losses. In large-scale farming, traditional disease detection techniques that primarily rely on expert inspection are frequently unfeasible. Deep learning and super-resolution methods for plant disease detection are thoroughly reviewed in this study, with an emphasis on their use in leaf-level imaging and UAV-based monitoring. This thorough review was carried out using the PRISMA framework, looking at peer-reviewed publications from important databases like Google Scholar, ScienceDirect, Web of Science, and Scopus. According to the review, low spatial resolution, environmental variability, occlusion, and domain shift cause convolutional and transformer-based models to perform poorly at canopy and field scales, despite achieving high accuracy on leaf-level datasets. Despite improving the perceptual quality of aerial imagery, super-resolution techniques are still difficult to incorporate into disease detection pipelines because of their high computational overhead, lack of task-aware training, scarcity of annotated UAV datasets, and poor generalization in real-world scenarios. There are few actual architectural implementations of cross-scale integration strategies currently in use; most of them are conceptual. Deep learning for plant disease detection has come a long way, but reliably deploying this technology outside of controlled environments remains challenging due to scale differences and data availability constraints. Coordinated developments in cross-scale learning approaches, data collection, and model design are needed to address these issues. The results also emphasize the significance of multimodal data fusion, super-resolution-assisted domain adaptation, and hierarchical transfer learning as viable approaches to enhancing the scalability and dependability of plant health monitoring systems. This review highlights useful research directions and provides a critical overview of current methods. It encourages the development of AI-powered solutions for better crop management, early disease detection, and sustainable farming methods.

Complex networks play a vital role in various real-world systems, including marketing, information dissemination, transportation, biological systems, and epidemic modeling. Identifying influential nodes within these networks is essential for optimizing spreading processes, controlling rumors, and preventing disease outbreaks. However, existing state-of-the-art methods for identifying influential nodes face notable limitations. For instance, Degree Centrality (DC) measures fail to account for global information, the K-shell method does not assign a unique ranking to nodes, and global measures are often computationally intensive. To overcome these challenges, this paper proposes a novel approach called Entropy Degree Distance Combination (EDDC), which integrates both local and global measures, such as degree, entropy, and distance. This approach incorporates local structure information by using entropy as a local metric and enhances the understanding of the overall graph structure by including path information as part of the global measure. This innovative method makes a substantial contribution to various applications, including virus spread modeling, viral marketing etc. The proposed approach is evaluated on six different benchmark datasets using well-known evaluation metrics and proved its efficiency.