MIND-SBERT: An Explainable and Trustworthy Article Retrieval and Summarization System

This paper presents MIND-SBERT, an innovative text analytics platform that transforms large-scale unstructured news data into actionable insights through an integrated analytical pipeline. The platform leverages the Microsoft News Dataset (MIND) and advanced natural language processing techniques to create a comprehensive analytics solution focused on transparency and trust. By combining Sentence-BERT (SBERT) for semantic analysis, GPT-4o for content synthesis, and a multi-dimensional analytical evaluation framework, MIND-SBERT provides a powerful tool for data-driven decision support. The platform incorporates detailed analytical performance metrics including quantitative measures (ROUGE, BLEU, METEOR) and qualitative dimensions (conciseness, relevance, coherence, accuracy, fairness) for comprehensive output assessment. A critical innovation is the trust auditing mechanism that identifies and corrects potential inaccuracies in analytical outputs. The interactive Streamlit-based analytics dashboard enables users to explore data patterns, analyze multi-document insights with transparent evaluation metrics, and understand the analytical process through detailed explanations. This human-centered analytics approach provides users with deep visibility into the platform's analytical decision-making, fostering trust and enabling informed use of generated insights for evidence-based decision support.