Higher-dimension time-series mining with manifold learning approach

Patent is one of the most important carriers of product innovation which provides richer technology information. Patent mining has significant effects for product innovation. Patent information can be act as higher-dimension time-series for it has the characteristics of time and higher-dimension. In this paper, we improved the locally linear embedding algorithm of manifold learning method. Then the patent can be transformed into a lower-dimension feature space. Experiment results show that after the transform process, the target patents would have the correlation. Our works would benefit a further patent mining research.