A Unified and Deployable Framework for Comparative Evaluation of Extractive, Abstractive, and LLM Based Text Summarization

Large scale unstructured text data used in enterprise systems, job markets, and online communication channels has enhanced the need to develop effective text summarization methods. Classical summarization models fall into extractive and abstractive, whereas more recent models based on large language models (LLM) offer significant context-sensitive and highly fluent summarization functions. Although research has been made in each paradigm, the majority of the studies contrast these methods and are mostly in controlled offline settings. This diversity reduces the possibility of a systematic comparison of the summarization paradigms and their adequacy in the real-world application. A deployable framework of the comparative assessment of extractive, abstractive, and LLM-based text summarization systems is proposed in this paper. The structure combines several models of summarization in a stratified architecture and compares them with each other through conventional metrics and statistical analysis. The system proposed is oriented to real-world applications like recruitment and resume analysis, where the summarization of the profiles of the candidates is essential. By using the regular evaluation metrics and comparative statistical analysis, the framework will seek to offer reproducible and deployment-oriented insights on the performance of various summarization paradigms.