TurkSentGraphExp: an inherent graph aware explainability framework from pre-trained LLM for Turkish sentiment...

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Computational Linguistics

Background:

• Sentiment classification is crucial for applications like customer feedback analysis and social media monitoring.
• Existing solutions often rely on black-box models, limiting explainability, especially for agglutinative languages like Turkish.
• Current explainability methods in Turkish NLP often fail to capture complex lexical and semantic relations.

Purpose of the Study:

• To propose a graph-aware explainability solution for Turkish sentiment analysis.
• To address the limitations of black-box models and improve the explainability of sentiment classification.
• To effectively handle the agglutinative nature of the Turkish language in sentiment analysis.

Main Methods:

• Development of TurkSentGraphExp, a graph-aware explainability framework for Turkish sentiment analysis.
• Utilizing graph representation learning (GRL) to capture semantic structures and relationships in Turkish text.
• Considering the semantic structure of suffixes and the agglutinative nature of Turkish.

Main Results:

• TurkSentGraphExp achieves a 10-40% improvement in explainability compared to state-of-the-art methods.
• The framework demonstrates robustness across varying sparsity levels.
• Case studies confirm the model's ability to identify semantic relationships from affixes in Turkish texts.

Conclusions:

• TurkSentGraphExp offers enhanced explainability for Turkish sentiment analysis by leveraging graph representations.
• The solution effectively captures the complex agglutinative structure of Turkish, providing rational phrase-level explainability.
• This work advances explainable AI in Turkish NLP by integrating semantic and structural information.