Feature Extraction and Classification of Text Data by Combining Two-stage Feature Selection Algorithm and Improved Machine Learning Algorithm

Abstract

The generation of massive text data makes efficient text classification a key requirement for information processing. However, the redundancy and uneven distribution of text data often lead to poor classification performance. To solve this problem, a two-stage feature selection algorithm using the fusion of information gain and maximum correlation minimum redundancy algorithm is proposed. To enhance the SVM performance in text data classification, an improved SVM algorithm based on Fourier hybrid kernel function was proposed. The results showed that on the IMDB dataset, the number of feature subsets required to achieve an accuracy of 0.82 for the proposed improved algorithm was only 40. When the number of features exceeded 390, the F1 value of the proposed algorithm was still about 1% to 2% higher than that of other algorithms. When the feature dimension was about 400, the improved algorithm proposed in this study performed best. The combination of the Fourier hybrid kernel function and the two-stage feature selection algorithm based on the information gain and maximum correlation minimum redundancy algorithm was 1%~3% higher in F1 value, and the number of correctly classified texts increased by 20 to 45. It can be found that the proposed algorithm provides an effective classification tool for processing large-scale text data, significant for information retrieval and data mining.