关键词: 时间序列, 半监督, 方差, 聚类

Abstract: Time series similarity is the important research direction of time series data mining.It is significant that how to make use of time series similarity to improve clustering of time series data.This paper presents a time series similarity-based semi-supervised spectral clustering algorithm.By selecting the appropriate features of time series to construct similarity and distance，the initial class is selected using tag data based on the spectral clustering algorithm.Results show the algorithm that makes time series with similar characteristics can be very effective to the same class are clustered.

Key words: time-series, semi-supervised, variance, clustering