An enhanced algorithm for semantic-based feature reduction in spam filtering.

Area of Science:

• Natural Language Processing
• Machine Learning
• Information Retrieval

Background:

• Synset-based text representations are popular for classification tasks.
• Ontological dictionaries like WordNet and BabelNet enhance these representations.
• Previous methods like Semantic Dimensionality Reduction System (SDRS) reduce dimensionality by combining semantically related features.

Purpose of the Study:

• To develop a novel lossless feature reduction scheme for synset-based text representations.
• To improve classification accuracy, particularly reducing false positive errors.
• To decrease the computational resources required compared to existing evolutionary algorithms.

Main Methods:

• Exploiting information from ontological dictionaries for feature reduction.
• A new lossless scheme that combines synsets based on class homogeneity in training data.
• Experimental validation on three datasets comparing against two optimization-based approaches.

Main Results:

• The proposed method achieves slightly better accuracy than optimization-based approaches, especially in reducing false positive errors.
• The new scheme significantly reduces computational resource requirements.
• Demonstrates effectiveness across multiple datasets.

Conclusions:

• The developed lossless feature reduction scheme offers a computationally efficient and accurate alternative to evolutionary algorithms.
• This approach effectively leverages ontological dictionaries for improved text classification.
• It provides a promising direction for dimensionality reduction in synset-based text representations.