Neuro-fuzzy network incorporating multiple lexicons for social sentiment analysis.

Area of Science:

• Artificial Intelligence
• Computational Linguistics
• Social Media Analytics

Background:

• Sentiment analysis of social media is challenging due to natural language ambiguity.
• Existing methods may not fully capture the nuances of user-generated content.
• A need exists for robust systems to assess content neutrality on social platforms.

Purpose of the Study:

• To introduce MultiLexANFIS, an adaptive neuro-fuzzy inference system (ANFIS) for classifying social media posts.
• To leverage multiple sentiment lexicons (VADER, AFINN, SentiWordNet) for enhanced sentiment analysis.
• To develop a system capable of distinguishing neutral from non-neutral (positive/negative) tweets.

Main Methods:

• Feature extraction integrating natural language processing (NLP) with fuzzy logic.
• Development of a novel set of 64 domain-independent fuzzy rules for the neuro-fuzzy network.
• Optimization of ANFIS parameters using gradient descent and least squares estimation.
• Utilizing sentiment scores from VADER, AFINN, and SentiWordNet as inputs.

Main Results:

• MultiLexANFIS achieved superior performance in classifying tweets into neutral and non-neutral categories.
• The system effectively handles the inherent fuzziness of natural language.
• Demonstrated effectiveness across various benchmark datasets compared to existing methods.
• Proposed single-lexicon ANFIS variants offer alternatives when multiple lexicons are unavailable.

Conclusions:

• MultiLexANFIS represents a significant advancement in social media sentiment analysis.
• The neuro-fuzzy approach offers a robust and efficient method for content neutrality assessment.
• The system's domain independence allows for broad applicability to diverse textual data.
• This research highlights the potential of integrating fuzzy logic and multiple lexicons for nuanced sentiment classification.