Explaining sentiment analysis results on social media texts through visualization.

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Computational Linguistics

Background:

• Artificial Intelligence (AI) systems achieve high performance but often lack transparency, risking bias propagation.
• Bias in AI, especially in widespread applications, can lead to significant societal damage.
• Interpretability is crucial for addressing fairness, accountability, and transparency in AI.

Purpose of the Study:

• To propose and evaluate two unique methods for enhancing AI decision-making interpretability.
• To implement and test these methods on Natural Language Processing (NLP)-based sentiment analysis.
• To improve the comprehensibility and trustworthiness of AI models in real-world applications.

Main Methods:

• Utilized Valence Aware Dictionary for Sentiment Reasoning (VADER) to generate heatmaps for visual justification.
• Employed Locally Interpretable Model-Agnostic Explanations (LIME) for in-depth analysis of model predictions.
• Applied these techniques to sentiment analysis tasks using social media data from Twitter, Facebook, and Reddit.

Main Results:

• The proposed visualization methods significantly increase the comprehensibility of sentiment analysis results.
• Heatmaps provide visual justification, aiding in understanding the model's reasoning process.
• LIME offers detailed insights into individual predictions, enhancing model transparency.

Conclusions:

• The developed interpretability techniques demonstrably improve understanding of AI decisions in sentiment analysis.
• The proposed system experimentally surpasses contemporary methods designed for AI interpretability.
• This work contributes to building more trustworthy and accountable AI systems through enhanced visualization.