Research On GPU Based Text Classification Algorithms

With the popularization and development of information technology on the Internet, the amount of Web pages increases exponentially. For people to get the information they want accurately in a short time, the Internet information must be classified and sorted in advance. Since most of the information of the Web pages is in text form, so the text classification has become a very important research topic in information technology field.Text classification is a process of making the texts in some haphazard order into one or several pre-defined categories according to its content automatically. The KNN (K-Nearest Neighbor) and SVM (Support Vector Machine) algorithms are the most widely used text classification algorithms. Although some researchers have improved them and made the categorization efficiency increased, as some reasons of the algorithms themselves, the computational complexities of the two classification algorithms are still very high.With the emergence and development of the CUDA architecture, high performance General-purpose computing based on graphics processors (GPU) has become more mature and sophisticated. To overcome the shortcomings of KNN and SVM text classification algorithms of high computational intensity, this thesis presents the GPU based text classification algorithms. The specific tasks are as follows:First of all, I put KNN algorithm was implemented on the CPU, then, at the text similarity computation and sorting stage of the KNN algorithm, a new parallel computing model is proposed based GPU; Finally, throuth understanding LIBSVM tool, the text training algorithm SMO (Sequential Minimal Optimization) of the SVM algorithm is accelerated with GPU parallel computing. Experiments prove that the proposed algorithms can greatly improve the efficiency of the KNN and SVM text classification algorithms with the quality of the classification being guaranteed.