Year: 2024
Type: conference-paper
Abstract: Uncovering the potential value of streaming data is not only critical for data analysis but also provides a strong foundation for data pricing and streaming data trend prediction.Compared to traditional batch mode clustering, there are two key challenges in stream clustering: (i) How to update the clustering results efficiently in the presence of constantly changing stream data?(ii) How to capture the evolutionary activities of clusters as they evolve with the data?However, most of the existing stream clustering algorithms cannot update the clustering results in real-time or track the evolution of clusters.The dynamic evolution of classes in streaming data reflects the real-time value and potential impact of the data.By accurately capturing these changes, we can gain a deeper understanding of the intrinsic value of the data.In response to this, we propose a stream evolution algorithm ECKStream by exploring the evolution of class kernel that accurately captures six evolutionary types of class within streaming data.Experiments show that ECKStream outperforms existing algorithms, providing high-quality clustering representations and effectively capturing the evolutionary patterns of different classes within streaming data.
Authors: Zhaohui Zhang, Dongxue Zhang, Ke Liu, Pei Zhang, Pengwei Wang
Institution: Donghua University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4406255380