Research of Spectral Clustering on Trend of Big Time Series

Automatic discrimination of big time series trends is researched in this paper. We focus on the three steps of traditional spectral clustering, that are similarity matrix construction, eigenvalue decomposition and eigenvector selection, and K-means clustering of the selected eigenvector, an algorithm based on MapReduce framework is designed. In order to achieve the goal for automatic identification of trends, we used the method of manually specifying the center point to guide the rest of the data to cluster into six typical trends. Based on the real data of Chinese stock market, the algorithm is proved to be feasible and effective.