Tag
This paper introduces Reference Abstraction (RA), Summary Abstraction (SA), and Abstraction Ratio (AR) metrics to quantify abstractiveness in text summarization, using harmonic mean of document lengths and cubic non-overlap factor. Empirical evaluation on XSUM with four models shows the metrics effectively discriminate between extractive and abstractive summaries, and flag potential hallucination.
This paper presents fine-tuning of PEGASUS on the XL-Sum English corpus, achieving state-of-the-art results with significant improvements over the baseline mT5 model across ROUGE scores.
ScholarSum is a hierarchical reflective graph-based framework for scientific abstractive summarization that emulates a student–teacher writing process. It uses a hierarchical knowledge graph to capture global structure, generates an initial draft, and iteratively refines it via evidence retrieval and teacher-like review to improve both fluency and factual faithfulness.
This paper extends optimal transport-based hallucination detection to all decoder layers in NMT and abstractive summarization, finding that detection is concentrated in early layers and that the geometric signal transfers poorly to summarization due to faithfulness failures not detectable via attention concentration.
Presents MASF, a multi-model adaptive selection framework that integrates multiple fine-tuned transformer summarization models and selects the highest-quality summary, achieving 88.63% BERTScore on CNN/DailyMail and outperforming several LLMs.