Decomposition and reduction of indexing structures with use of the GPU computations

Abstract

The purpose of this article is to determine the usefulness of the K-means method applied to decompose and reduce the TERM-BY-DOCUMENT matrix with use of the GPU computations. The reduction based on the truncated SVD decomposition is a long-term process for large matrices. The computational complexity of the SVD decomposition is O(n3). The use of the K-means decomposition before reduction should reduce the computational complexity. In addition, the use of the GPU should reduce the time of the whole process. In the article there is comparison of the times and the correctness of the results for the computing environments associated with the GPU.