With A Knowledge Adaption Mechanism: A Novel Horizontal Collaborative Fuzzy Clustering Method

Being one of the most important models for multisource data clustering, Horizontal Collaborative Fuzzy Clustering (HCFC) implements the clustering by fusing the knowledge obtained from external source datasets during the clustering on local source dataset. The external knowledge obtained from each given external source dataset is represented by a partition matrix for this dataset, revealing the data structure from the corresponding perspective. Meanwhile, the sizes of the partition matrixes for external source datasets and local source dataset are required to be equal. This prerequisite is not always guaranteed in real applications due to the different structures of external datasets. In such cases, the HCFC algorithm fails to work, resulting in a limited application scope. Aim at this defect, this paper presents a novel Knowledge Adaption algorithm based on Spectral clustering and Relative entropy (KASR) to deal with this issue. Based on the proposed KASR, a novel algorithm called Horizontal Collaborative Fuzzy Clustering with Knowledge Adaption mechanism (HCFCKA) is formulated, which can work well when the sizes of partition matrixes for external source datasets and local source dataset are different and thus extends the application scope of the HCFC algorithm. Experiments on synthetic dataset and UCI real-world datasets demonstrate that the proposed KASR outperforms the existing knowledge adaption methods and the proposed HCFCKA can achieve significantly better collaborative clustering performance compared with the state-of-art methods.