Hybrid Extractive Text Summarization: Combining TF-IDF and TextRank for Factual and Explainable Summaries

This thesis presents a hybrid extractive text summarization system that synergistically combines Term Frequency-Inverse Document Frequency (TF-IDF) weighting with the graph-based TextRank algorithm. The system is designed to address the critical need for factual integrity and explainability in automated summarization, particularly in high-stakes domains such as academic research and technical documentation. By leveraging the lexical precision of TF-IDF for initial node weighting and the structural intelligence of graph centrality for ranking, the proposed methodology overcomes the "contextual blindness" of purely statistical methods and the "uniform initialization" limitations of standard graph-based approaches. The system features a novel dual-format output—providing both narrative paragraphs and structured bullet points—to accommodate diverse user cognitive modes. Implemented as a privacy-preserving, client-side web application, the system requires no training data or external API dependencies. Evaluation against benchmark datasets demonstrates competitive performance with state-of-the-art extractive methods while maintaining complete transparency and computational efficiency.