A linear-RBF multikernel SVM to classify big text corpora.

Area of Science:

• Machine Learning
• Computational Linguistics
• Data Mining

Background:

• Support Vector Machines (SVM) are effective for classification but struggle with large datasets due to high training complexity.
• Existing kernel methods highlight the need for multiple kernels or flexible parameterizations for enhanced performance.
• Handling high-dimensional data and text corpora with SVMs remains a significant challenge in machine learning.

Purpose of the Study:

• To introduce a novel multikernel Support Vector Machine (SVM) designed for efficient classification of high-dimensional data.
• To develop an automatic parameterization method for multikernel SVMs that reduces computational cost.
• To improve classification accuracy and training speed compared to traditional SVMs and brute-force parameterization techniques.

Main Methods:

• The proposed model structures data by spreading it into cohesive term slices (clusters) to form a multikernel.
• Automatic parameterization is employed to optimize kernel configurations.
• The multikernel SVM approach is evaluated on diverse text corpora.

Main Results:

• The multikernel SVM demonstrates good classification accuracy on various text corpora.
• The new approach achieves significantly faster training times compared to classic SVM classifiers.
• Experimental results show superior performance against SVMs with brute-force parameterization.

Conclusions:

• The developed multikernel SVM effectively manages high-dimensional data and text classification tasks.
• This method offers a computationally efficient and accurate alternative to conventional SVMs.
• The findings suggest a promising direction for scaling SVMs to larger and more complex datasets.