摘要

Large-scale data are any data that cannot be loaded into the main memory of the ordinary. This is not the objective definition of large-scale data, but it is easy to understand what the large-scale data is. We first introduce some present algorithms to clustering large-scale data, some data stream clustering algorithms based on FCM algorithms are also introduced. In this paper, we propose a new structure to cluster large-scale data and two new data stream clustering algorithms based on the structure are propose in Sects. 3 and 4. In our method, we load the objects in the dataset one by one. We set a threshold of the membership, if the membership of one object and a cluster center is bigger than the threshold, the object is assigned to the cluster and the location of nearest cluster center will be updated, else the object is put into the temporary matrix; we call it pool. When the pool is full, we cluster the data in the pool and update the location of cluster centers. The two algorithms are based on the data stream structure. The difference of the two algorithms is the how the objects in the data are weighed. We test our algorithms on handwritten digits images dataset and several large-scale UCI datasets and make a comparison with some presented algorithms. The experiments proved that our algorithm is more suitable to cluster large-scale datasets.