A Novel Parallel LSA-SVM Algorithm Based on Semantic Distance for Blog

Abstract

Emotional analysis can be considered as a kind of classification of sentiment polarity in essence. Against the background of mass data processing, in order to increase the accuracy of judgment on the emotion conveyed by a text, a method to classify the emotional tendency of a text that combines Latent Semantic Analysis (LSA) and Support Vector Machine (SVM) is proposed herein. By this method, a semantic distance vector space modal of “word-document” is developed from semantic aspect following the method of LSA. Then, with the help of SVM that is featured by high classification accuracy and good generalization ability, the emotion is classified. At last, this paper proposed a parallel implementation of LSA-SVM algorithm. The algorithm is developed using Message Passing Interface (MPI) in parallel environment. Experiments show that the accuracy of this method is higher than that of the conventional SVM method in the Blog assessment where sentences are short and emotional tendency is evident, the classification accuracy in a test set approximates to 92.2%, and compared with the serial implementation, the parallel LSA-SVM algorithm increases efficiency significantly.