Abstract: Linear Neighborhood Propagation（LNP） is a very effectively graph-based semi-supervised classification method. However, different class overlapping and data distributed imbalance cause the choice of the neighbors to be unreasonable when constructing the graph in LNP. This paper applies spectral clustering to analyzing the data’s distribution, adjusts the distance metric in the choice of neighbors to make the neighbors more reasonable according to the clustering result. LNP method based on spectral clustering is applied to the time series classification. Using four time series datasets from UCR time series data mining archive, the experimental results show that spectral clustering based LNP acquires higher accuracy than LNP.

Key words: Linear Neighborhood Propagation（LNP）, semi-supervised, class overlapping, imbalance

摘要： 线性邻近点传播（LNP）是一种非常有效的基于图的半监督分类方法，而类重叠与数据分布不平衡问题会使LNP构造图时由于选择的邻居不合理而影响分类性能。采用谱聚类来分析数据的分布，根据聚类结果对邻居选择时的距离度量进行调整，使得选择的邻居更合理。将基于谱聚类的LNP方法应用于时间序列分类，在UCR时间序列挖掘库的四个数据集上进行实验，结果表明该方法比LNP方法具有更高的分类准确率。

关键词: 线性邻近点传播, 半监督, 类重叠, 分布不平衡