Interpretable Text Embeddings and Text Similarity Explanation: A Primer

2025-02-20

AI Summary

Similarity scores derived from text embeddings lack interpretability, hindering their trustworthy deployment in transparency-critical NLP applications such as search. Method: This paper establishes the first structured methodology framework for explaining text similarity, proposing a unified taxonomy and a multidimensional evaluation framework—encompassing faithfulness, readability, and efficiency. We systematically survey and empirically evaluate five mainstream explanation paradigms: attention attribution, perturbation-based analysis, prototype learning, token-level importance mapping, and generative explanation. Contribution/Results: Our analysis reveals inherent trade-offs between explanation quality and computational overhead across methods and clarifies their respective applicability boundaries. The work provides a foundational theoretical paradigm for explainable text embedding research and delivers empirically grounded guidelines for industrial practitioners to select and customize explanation solutions according to task-specific requirements.

Technology Category

• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models
• Computer Vision: Interpretability, Explainability, and Transparency
• Machine Learning: Transparent, Interpretable, Explainable ML

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Text embeddings and text embedding models are a backbone of many AI and NLP systems, particularly those involving search. However, interpretability challenges persist, especially in explaining obtained similarity scores, which is crucial for applications requiring transparency. In this paper, we give a structured overview of interpretability methods specializing in explaining those similarity scores, an emerging research area. We study the methods' individual ideas and techniques, evaluating their potential for improving interpretability of text embeddings and explaining predicted similarities.

Problem

• Explaining text similarity scores
• Improving interpretability of text embeddings
• Structured overview of interpretability methods

Innovation

• Interpretable text embeddings
• Explain text similarity scores
• Evaluate interpretability methods