Text Feature Extraction for Public English Vocabulary Based on Wavelet Transform

Area of Science:

• Natural Language Processing
• Signal Processing
• Machine Learning

Background:

• Text feature extraction is crucial for computers to understand natural language.
• High-dimensional text data often suffers from low feature differentiation.
• Traditional methods like TF-IDF face challenges with complex text data.

Purpose of the Study:

• To propose an improved text feature extraction method using wavelet analysis.
• To address the frequency aliasing issue in wavelet-based signal decomposition.
• To enhance feature differentiation and reduce dimensionality in text data.

Main Methods:

• Applied fast discrete wavelet transform and inverse discrete wavelet transform to TF-IDF vectors.
• Developed an improved inverse discrete wavelet transform to mitigate frequency aliasing.
• Utilized reconstructed wavelet coefficients for signal analysis at each scale.

Main Results:

• The proposed wavelet transform-based method outperforms existing feature extraction techniques.
• Achieved higher classification accuracy on public English vocabulary datasets.
• Successfully reduced the dimensionality of the TF-IDF vector space model.

Conclusions:

• The improved wavelet transform method offers a robust solution for text feature extraction.
• This approach enhances the effectiveness of natural language processing tasks.
• The method provides a balance between accuracy and dimensionality reduction.