Article Preview
Top1. Introduction
At present, with the rapid development of information technology such as computers, networks, and communications, information processing is rapidly industrializing on the entire social scale (Wilson et al., 2015; Zheng, 2015). In particular, the rapid rise of platforms such as “Internet +” has put people in the “ocean” of data all the time (Liu, 2019). Behind these huge amounts of data often contains very rich information. Relying on traditional data retrieval and analysis methods can no longer solve our need to obtain information. Traditional data management models are no longer suitable for today's data management needs. Therefore, data mining technology, the rapid rise has become one of the efficient solutions to how to quickly obtain useful information in today's society (Shu et al., 2017).
Data mining is to extract valid and potentially useful information from a large amount of raw data, and then express it through an understandable model (Chen, Feng, & Mao, 2019). Therefore, data mining has become more and more important and favored. Data mining is both a data processing method and a data processing process, and cluster analysis is an important method in data mining. K-means algorithm is one of the most classic and simple methods in cluster analysis. Although the K-means algorithm is simple to operate and easy to understand, there are some problems, such as ambiguity and subjective experience in determining the value of k, and the random selection of k initial center points has a great impact on the clustering result.
The text data clustering research has been developed for a long time since it was proposed, and its related technical methods have been initially formed, but the existing technical methods have more or less certain deficiencies and defects. In recent years, with the continuous deepening of data mining technology, more and more neighborhoods can see the application of k-means algorithm. Diab Abuaiadah (Abuaiadah, 2016) studied the analysis of five commonly used similarity and distance functions (Pearson's correlation coefficient, cosine, Jacquard coefficient, Euclidean distance, and average Kullback-Leibler divergence) when analyzing Arabic documents. The performance of the bisect k-means clustering algorithm and the traditional k-means algorithm are presented. The Lu team (Lu & Lu, 2016) proposed a fast genetic K-means algorithm (FGKA). The algorithm is improved on the GKA algorithm. Experiments show that FGKA and GKA always always converge to the global optimum, and that FGKA runs much faster than GKA. The Mexicano A team (Mexicano et al., 2015) proposed a fast mean algorithm based on the K-means algorithm, which can reduce the transaction data set time by up to 99.02%, while reducing the quality by only 7.62%. The Chunhua P team (Chunhua et al., 2015) aimed at the fluctuation of DG (distributed power generation) output and the uncertainty of load demand, which may cause the planned DG capacity to become larger or the improvement rate of the voltage curve to be reduced. K-means clustering and multi-scenario probability analysis are proposed and used to reduce the impact of volatility and uncertainty on the distribution network. The Dimitri J team (Dimitri, 2018) applied the k-means algorithm to the latest commercial mixed integer programming solver and achieved significant improvements. Wang Y's team (Wang & Deng, 2015) has some limitations on HMAX models in modeling V2 neurons or higher-level visual cortex during the simulation of visual recognition in primate visual cortex. The k-means algorithm is proposed to be applied to the model. The experimental results show that the method effectively solves these limitations. The Niu B team (Niu et al., 2016) proposed to mix five PSO and K-Means, which have shown good prospects in continuous function optimization, respectively, and thus obtained five PSO-KM-based clustering methods.