Research on performance variations of classifiers with the influence of pre-processing methods for Chinese short text...

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Machine Learning

Background:

• Text preprocessing is crucial for Chinese text classification.
• Existing research primarily focuses on English text preprocessing.
• Limited studies explore preprocessing's impact on diverse Chinese text classification algorithms.

Purpose of the Study:

• To experimentally compare Chinese text preprocessing methods.
• To evaluate their influence on fifteen common text classifiers.
• To analyze performance under various conditions like evaluation metrics and classifier types.

Main Methods:

• Utilized three standard Chinese preprocessing techniques: word segmentation, stop word removal, and symbol removal.
• Tested fifteen classifiers on two Chinese datasets.
• Analyzed classification results using metrics like macro-F1, considering different preprocessing combinations and classifier selections.

Main Results:

• Most classifiers showed improved performance after applying appropriate preprocessing.
• Systematic preprocessing positively impacted Chinese short text classification.
• Achieved macro-F1 scores of 92.13% and 91.99%, outperforming baselines by 0.3% and 2% respectively.

Conclusions:

• Systematic application of preprocessing methods enhances Chinese short text classification.
• Word segmentation, stop word removal, and symbol removal are effective preprocessing steps.
• Preprocessing benefits both machine learning and deep learning models.