The Research On Svm-based Method Of Information Classification In Pervasive Computing Applications

In the pervasive computing environment, the context information analysis is significantly important because it is vital in deciding whether applications can provide proper service. The pervasive applications demand the rapid and efficient classification and management in pervasive environment. However, the information collection is often of large scale and various categories. As a result, to find an effective method to classify and manage the information in pervasive computing environment becomes more meaningful.The support vector machine is a kind of machine learning method based on statistic learning theory which shows efficiency in nonlinear and high dimension training. SVM has received great quality in text-categorizing applications and is widely used in facial recognition, image processing and other research fields.Based on the research of many SVM algorithms, this paper proposes a new semi-sparse approach for vector multiplication and applies it to the Sequential Minimal Optimization algorithm, which improves the speed of vector multiplication in large scale sparse matrix and enhances the performance of the SVMTorch classifier. Theoretical analysis indicates that the traditional sparse method will spend O(m+n) time on the comparing and addressing of two vectors with m and n values. However, the new semi-sparse algorithm only costs O(n) time to finish the multiplication without reducing the accuracy of SVM classifier. The experiments also show that the efficiency of semi-sparse-based SVMTorch classifier is much better than the original classifier. The training process based on two implementations of the semi-sparse algorithm has been reduced to 54.32% and 74.95% on the time consumption compared to the sparse algorithm.Meanwhile, this paper extends the SVMTorch classifier which is now supporting both the single-labeled classification and the multi-labeled classification. The training and testing of multi-labeled classification of improved SVMTorch are verified with Reuters-21578 dataset.In order to improve the performance of SVMTorch classifier, this paper adopts MPI to parallelize it and now it runswell on multi-core processors and clusters. Finally a parallel SVMTorch model supporting multi-labeled classification based on semi-sparse algorithm is implemented and applyed on Chinese webpage classification.