Jul 2026· Recent Advances in Computer Science and Communications· 0 citations
Abstract
The growing volume of scientific literature has driven the demand for
automated text summarization systems that are natural and factual. Extractive Text
Summarization methods are factually accurate in meaning; still, they can lead to a summary
that is not cohesive. On the other hand, abstractive summarization systems improve readability
but may introduce factual bias. The paper overcomes these shortcomings by creating a hybrid
text summarization system that combines extractive and abstractive methods to maximize both
quality and factual content.
The framework uses two unsupervised extractive models, HipoRank and PacSum, to
extract important sentences, which are then synthesized with the original input document's
introduction section and subjected to long-document transformer models, PEGASUSX and
LED, to generate abstract-style summaries.
Among the tested combinations, the HipoRank-LED configuration achieved the most balanced performance, with ROUGE-1: 0.440, ROUGE-2: 0.220, and ROUGE-L: 0.410 on the PubMed dataset. This combination occasionally produced summaries with greater abstractiveness than the human-written references.
Various experiments across the ScisummNet, ArXiv, and PubMed datasets show that
hybrid configurations are always better than extractive and abstractive ones. HipoRank-LED is
the most efficient model, with ROUGE-1 = 0.440, ROUGE-2 = 0.220, and ROUGE-L = 0.410
on PubMed.
Results indicate that combining extractive grounding with long-context
transformers improves informativeness and coherence and reduces hallucination errors. The
introduction-guided structured input also provides better global context for summarizing
complex scientific documents.
The findings indicate that the transformer-based abstraction, combined with an
extractive text summarization approach, can be a very useful, scalable, and domain-independent
model for approximating long scientific texts.
The rapid growth of academic literature has intensified the need for effective automatic text summarization techniques capable of producing concise and informative representations of scientific documents. While extractive methods are widely used, they are limited in their ability to generate coherent and semantically rich summaries. Recent advances in Transformer-based architectures have enabled significant progress in abstractive summarization; however, their effectiveness on domain-specific datasets, such as scientific articles, remains an open challenge. In this study, we investigate the performance of three pre-trained Transformer-based models—T5, BART, and GPT-2—on the task of abstractive summarization using the CL-SciSumm 2019 dataset. A total of 19 experimental configurations are conducted to analyze the impact of generation parameters, including beam size, length penalties, and n-gram constraints, on summarization quality. The models are evaluated using ROUGE metrics, with a focus on ROUGE-2.To complement content-based evaluation, this work incorporates linguistic acceptability assessment using the Corpus of Linguistic Acceptability (CoLA), a benchmark dataset for evaluating grammatical correctness. The results show that BART achieves the best performance with an ROUGE-2 F1-score of 0.40664, while T5 demonstrates superior grammatical acceptability, achieving 93.36%, but BART achieves a very near performance to T5. Ultimately, these findings demonstrate the potential of pre-trained neural networks, particularly the BART architecture, to drive the future of complex, generative NLP applications, transforming how academic research is processed and understood.
Emad Nabil· Islamic University Journal o...· 0 citations
The exponential growth of scientific literature has intensified the demand for automated summarization systems capable of producing abstracts that are both linguistically fluent and factually reliable. Existing approaches face a fundamental trade-off: encoder-decoder models such as BART and T5 maintain strong factual grounding but produce rigid, extractive outputs, while decoder-only large language models (LLMs) such as Llama and Gemma generate highly fluent text yet remain susceptible to hallucination. This paper proposes a two-stage Synergistic Hybrid Ensemble framework designed to resolve this dichotomy. In Stage 1, a fine-tuned BART-Large model generates a factually grounded scaffold draft from a structured input representation comprising the document title, key sentences, method highlights, and results summary. In Stage 2, a QLoRA-adapted Llama-3.2-1B model performs coherent rewriting and stylistic polishing by conditioning on both the scaffold draft and the original source document. Experiments conducted on the arXiv Scientific Research Papers Dataset using BERTScore and entailment-based Factual Consistency metrics demonstrate that the proposed ensemble achieves a Factual Consistency metrics demonstrate that the proposed ensemble achieves a Factual Consistency score of 0.9140, substantially outperforming BART-Large (0.2890) and Llama-3.2-1B (0.6630) individually. Although the ensemble incurs a marginal reduction in BERTScore (0.8980) relative to Llama-3.2-1B (0.9555), this trade-off is justified given the critical importance of factual reliability in high-stakes scientific discourse. These findings confirm that anchoring the generative capacity of decoder-only LLMs to verified factual scaffolds effectively mitigates hallucination risk, offering a scalable and reproducible solution for high-fidelity scientific abstract generation.
Geoffrey Antonio Arifin, Andrew Widyanata, Henry Lucky et al.· International Conference on...· 0 citations
The increasing number of scientific publications has made literature screening more time-consuming, particularly for researchers who need to identify the main contribution of an article before reading the full text. This study develops an extractive summarization model for Indonesian scientific articles using a BERT-based transfer learning approach. The proposed method represents each sentence with contextual embeddings and selects relevant sentences based on their similarity to the document representation, while applying a redundancy threshold to reduce redundancy. A curated corpus of Indonesian research articles was used for model development and evaluation. The generated summaries were evaluated using ROUGE-1, ROUGE-2, ROUGE-L, and ROUGE-Lsum, with the article abstract used as the reference summary. The experimental results show that the proposed model achieved a ROUGE-1 score of approximately 33%, indicating it retained important keywords and central information from the source documents. However, the lower ROUGE-2 score suggests that the model still has limitations in preserving phrase-level continuity and sentence coherence. Qualitative analysis also shows that the model can capture the main ideas of scientific articles, although some methodological details and contextual information are occasionally omitted. These findings indicate that BERT-based extractive summarization can support preliminary literature screening, but further improvement is needed through stronger baseline comparison, human evaluation, and redundancy-aware optimization.
Made Hanindia Prami Swari, Puji Lestari Tarigan, Gusti Eka Yuliastuti et al.· JOIV: International Journal...· 0 citations
The developed application is an example of the effective use of Artificial Intelligence in processing documents in regional languages and lays the groundwork for creating automated systems for document management.
Apoorva S., Usha B. S., S. Darshan· International Journal of Inn...· 0 citations
The proposed framework using Bidirectional Long Short-Term Memory with a hypergraph and a dominating set mechanism proves to be an efficient approach to automatic summarization and has the potential to be applied in journalism, healthcare, legal analysis, and digital content management.
Pradeepa Sampath, S. Subashini, V. Shanmuganathan et al.· Artificial Intelligence and...· 0 citations
This survey presents a systematic review of 121 references spanning 2002 to 2026, tracing the evolution of TextRank-based approaches into hybrid LLM pipelines and advancing three qualified arguments.
Ahmed J. Jabur, Asmaa Abdul Azeez Dakhil, Israa Saad Mohammed et al.· Iraqi Journal for Computers...· 0 citations