Revolutionizing Chinese sentiment analysis: A knowledge-driven approach with multi-granularity semantic features

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Artificial Intelligence

Background:

• Chinese text sentiment analysis research has advanced, but gaps remain regarding cross-lingual differences, domain knowledge, and task-specific needs.
• Existing methods often overlook the unique linguistic characteristics of Chinese text, limiting sentiment analysis performance.
• Practical applications require more robust and accurate sentiment detection tailored to the nuances of Chinese language.

Purpose of the Study:

• To address the limitations in current Chinese text sentiment analysis.
• To propose a novel method that deeply integrates emotional knowledge and linguistic features.
• To enhance the accuracy and effectiveness of sentiment detection for Chinese texts.

Main Methods:

• Proposed a method integrating knowledge vectors from emotional knowledge triplets (using TransE) with BiGRU and attention mechanism feature vectors.
• Introduced radical and emotional part-of-speech features, leveraging character and word characteristics.
• Developed a collaborative approach integrating multi-granularity features: characters, words, radicals, and parts of speech.

Main Results:

• Achieved an F1-score of 89.23% on the Douban Film Review dataset.
• Achieved an F1-score of 84.84% on the NLPECC dataset.
• Demonstrated significant improvement in sentiment detection accuracy by leveraging emotional insights and linguistic elements.

Conclusions:

• The proposed method effectively leverages emotional knowledge and linguistic nuances for superior Chinese sentiment analysis.
• The integration of multi-granularity features significantly bolsters the accuracy of sentiment detection.
• The approach shows strong efficacy and potential for practical applications in Chinese sentiment analysis.