Automatic Construction and Global Optimization of a Multisentiment Lexicon.

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Sentiment Analysis

Background:

• Manual sentiment lexicon creation is labor-intensive, time-consuming, and lacks emotional intensity quantification.
• Domain-specific lexicons have limited applicability.
• Existing methods struggle with context-dependent word meanings.

Purpose of the Study:

• To develop an automatic method for constructing a multidimensional sentiment lexicon for Chinese.
• To enhance sentiment lexicon flexibility and accuracy through disambiguation and a global optimization framework.
• To introduce the 10-dimensional sentiment lexicon SentiRuc.

Main Methods:

• Statistical training on a large-scale Chinese corpus using a neural network language model.
• Construction of a multidimensional sentiment lexicon based on coordinate offset constraints.
• Development of a sentiment disambiguation algorithm and a global optimization framework for combining human-annotated resources.

Main Results:

• The SentiRuc lexicon demonstrates superior performance in category labeling, intensity labeling, and sentiment classification tasks.
• SentiRuc achieved a 21% lead over the next best method in the intensity label test.
• The proposed methods effectively address the limitations of manual and domain-specific sentiment lexicons.

Conclusions:

• SentiRuc offers an effective and efficient approach to building high-performance multidimensional sentiment lexicons.
• The developed methods enhance the accuracy and flexibility of sentiment analysis, particularly for Chinese text.
• This work contributes to advancing sentiment analysis research through automated lexicon construction.