Design and Application of a Text Clustering Algorithm Based on Parallelized K-Means Clustering

Abstract

The traditional text clustering algorithms face two common problems: the high dimensionality of computing vectors and poor calculation efficiency. To solve these problems, this paper explores deep into the K-means clustering (KMC), Hadoop and Spark big data technique, and then proposes a novel text clustering algorithm based on the KMC parallelized on big data platform. The propose algorithm is denoted as the SWCK-means. First, the Word2vec was adopted to calculate the weights of word vectors, and thus reduce the dimensionality of the massive text data. Next, the Canopy algorithm was introduced to cluster the weight data, and identify the initial cluster centers for the KMC. On this basis, the KMC was employed to cluster the preprocessed data. To improve the efficiency, a parallel design for the Canopy algorithm and the KMC was developed under the Spark architecture. The proposed algorithm was verified through experiments on a massive amount of online text data. The results show that our algorithm achieved more accurate classification effects than the traditional KMC, especially in handling a huge amount of data.