Text classification algorithm of tourist attractions subcategories with modified TF-IDF and Word2Vec

Area of Science:

• Natural Language Processing
• Text Mining
• Machine Learning

Background:

• Text classification is crucial for managing big data, but its application to specialized domains like tourist attractions is underexplored.
• Existing methods struggle with complex, imbalanced datasets common in professional fields.

Purpose of the Study:

• To develop and validate a novel text representation and classification method for tourist attraction descriptions.
• To address the challenges of classifying complex, imbalanced, and domain-specific text data.

Main Methods:

• Constructed a corpus of tourist attraction descriptions using web crawlers.
• Proposed a hybrid text representation combining Word2Vec embeddings with TF-IDF-CRF-POS weighting.
• Integrated the representation with seven common classifiers (DT, SVM, LR, NB, MLP, RF, KNN) for multi-class classification.

Main Results:

• The proposed algorithm achieved higher accuracy (2.29%) and F1-scores (macro-F1: 5.55%, micro-F1: 2.90%) compared to traditional methods and BERT.
• Demonstrated superior performance on imbalanced datasets, identifying rare categories effectively.
• Exhibited enhanced stability across datasets of varying sizes.

Conclusions:

• The novel text representation and classification algorithm offers superior performance and robustness for professional domain text.
• The method is practical and provides a valuable reference for vector expression and classification of complex Chinese text datasets.