PLSGA：阶段式长文本摘要生成方法

doi:10.3778/j.issn.1002-8331.2309-0048

摘要/Abstract

摘要： 针对现有方法在处理长文本时，存在冗余信息处理困难和无法筛选出最高质量摘要的问题，提出了一种阶段式长文本摘要生成方法（PLSGA）。将样本数据的文本和参考摘要分割，利用Sentence-BERT获取语义向量并进行相似度比对，从中抽取文本的关键信息；通过关键信息和非关键信息训练抽取模型，以尽可能地保留原文本的语义信息；将抽取的关键信息和参考摘要作为样本输入骨干模型BART进行生成模型训练；通过生成模型生成多条候选摘要，并使用无参考摘要评分模型筛选出质量最好的摘要。提出的阶段式长文本摘要生成方法在多个中文长文本数据集上进行实验，结果表明相对于目前主流的方法以及ChatGPT，其效果均有提升，具有领域优势，生成的摘要质量更好，更具可读性。

关键词: 文本摘要, Sentence-BERT, 关键信息, BART, 无参考摘要评分模型

Abstract: Aiming at the problem that the existing methods have difficulty in processing redundant information and cannot select the highest quality abstract when dealing with long text, this paper proposes a staged long text abstract generation method (PLSGA). Firstly, the paper segments the text of the sample data and the reference summary, and uses Sentence-BERT to compare and extract the key information of the text. The paper trains the extraction model through key information and non-key information to retain the semantic information of the original text as much as possible. The extracted key information and reference summaries are input as samples into the backbone model BART for generative model training. Finally, multiple candidate summaries are generated through the generative model, and the best-quality summaries are selected using the no-reference summaries scoring model. The experiment proves that the proposed stage-based long text summary generation method has been tested on multiple Chinese long text data sets. The results show that compared with the current mainstream method and ChatGPT, its effect has been improved, having domain advantages, and the quality of the generated summary is much better and more readable.

Key words: text summarization, Sentence-BERT, key information, BART, no-reference summarization scoring model

方缙, 李宝安, 游新冬, 吕学强. PLSGA：阶段式长文本摘要生成方法[J]. 计算机工程与应用, 2025, 61(2): 170-178.

FANG Jin, LI Bao’an, YOU Xindong, LYU Xueqiang. PLSGA： Phase-Wise Long Text Summary Generation Approach[J]. Computer Engineering and Applications, 2025, 61(2): 170-178.

参考文献

[1] 皮洲, 奚雪峰, 崔志明, 等.一种面向长文本小数据集自动摘要任务的数据增强策略[J].中文信息学报, 2022, 36(9): 46-56.
PI Z, Ⅺ X F, CUI Z M, et al. A data augmentation method for long text automatic summarization[J].Journal of Chinese Information Processing, 2022, 36(9): 46-56.
[2] 黄于欣, 余正涛, 郭军军, 等.基于主题交互图的案件话题摘要[J].软件学报, 2023, 34(4): 1796-1810.
HUANG Y X, YU Z T, GUO J J, et al.Case-related topic summarization based on topic interaction graph[J]. Journal of Software, 2023, 34(4): 1796-1810.
[3] 李刚, 余正涛, 黄于欣.案件要素异构图的舆情新闻抽取式摘要[J].计算机工程与应用, 2023, 59(4): 112-119.
LI G, YU Z T, HUANG Y X. Extractive summary for public opinion news via case elements heterogeneous graph[J]. Computer Engineering and Applications, 2023, 59(4): 112-119.
[4] 余帅, 宋玉梅, 秦永彬, 等.基于审判逻辑步骤的裁判文书摘要生成方法[J].计算机工程与应用, 2024, 60(4): 113-121.
YU S, SONG Y M, QIN Y B, et al. Method for generating summary of judgment documents based on trial logic steps[J]. Computer Engineering and Applications, 2024, 60(4): 113-121.
[5] SUTSKEVER I, VINYALS O, LE Q V. Sequence to sequence learning with neural networks[C]//Proceedings of the 27th International Conference on Neural Information Processing Systems, 2014: 3104-3112.
[6] REIMERS N, GUREVYCH I. Sentence-BERT: sentence embeddings using siamese BERT-networks[J]. arXiv:1908.
10084, 2019.
[7] DEVLIN J, CHANG M W, LEE K, et al. BERT: pre-training of deep bidirectional transformers for language understanding[J]. arXiv:1810.04805, 2018.
[8] VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems, 2017: 6000-6010.
[9] CER D, DIAB M, AGIRRE E, et al. SemEval-2017 task 1: semantic textual similarity-multilingual and cross-lingual focused evaluation[J]. arXiv:1708.00055, 2017.
[10] LIU Y, OTT M, GOYAL N, et al. RoBERTa: a robustly optimized BERT pretraining approach[J]. arXiv:1907.11692, 2019.
[11] KIROS R, ZHU Y, SALAKHUTDINOV R R, et al. Skip-thought vectors[J]. arXiv:1506.06726, 2015.
[12] WILLIAMS A, NANGIA N, BOWMAN S R.A broad-coverage challenge corpus for sentence understanding through inference[J]. arXiv:1704.05426, 2017.
[13] HILL F, CHO K, KORHONEN A. Learning distributed representations of sentences from unlabelled data[J]. arXiv: 1602.03483, 2016.
[14] HUMEAU S, SHUSTER K, LACHAUX M A, et al. Poly-encoders: transformer architectures and pre-training strategies for fast and accurate multi-sentence scoring[J]. arXiv: 1905.01969, 2019.
[15] MCKEOWN K, RADEV D R. Generating summaries of multiple news articles[C]//Proceedings of the 18th Annual International Conference on Research and Development in Information Retrieval, 1995: 74-82.
[16] CHOPRA S, AULI M, RUSH A M. Abstractive sentence summarization with attentive recurrent neural networks[C]//Proceedings of the Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2016: 93-98.
[17] LOPYREV K. Generating news headlines with recurrent neural networks[J]. arXiv: 1512.01712, 2015.
[18] BAHDANAU D, CHO K, BENGIO Y. Neural machine translation by jointly learning to align and translate[J].arXiv:1409.0473, 2014.
[19] LEWIS M, LIU Y, GOYAL N, et al. BART: denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension[J]. arXiv:1910.13461, 2019.
[20] RAFFEL C, SHAZEER N, ROBERTS A, et al. Exploring the limits of transfer learning with a unified text-to-text transformer[J]. The Journal of Machine Learning Research, 2020, 21(1): 5485-5551.
[21] LI W, GAO C, NIU G, et al.UNIMO: towards unified-modal understanding and generation via cross-modal contrastive learning[J]. arXiv:2012.15409, 2020.
[22] BAKRE M, SUN P, SHEN X, et al. HRS-bench: holistic, reliable and scalable benchmark for text-to-image models[J]. arXiv:2304.05390, 2023.
[23] 王宗辉, 李宝安, 吕学强, 等. BETES: 一种中文长文档抽取式摘要方法[J].小型微型计算机系统, 2022, 43(1): 42-49.
WANG Z H, LI B A, LV X Q, et al. Method of extractive summarization Chinese long documents[J].Journal of Chinese Computer Systems, 2022, 43(1): 42-49.
[24] 张乐, 冷基栋, 吕学强, 等. RLCPAR: 一种基于强化学习的中文专利摘要改写模型[J].数据分析与知识发现, 2021, 5(7): 59-69.
ZHANG L, LENG J D, LV X Q, et al. RLCPAR: a rewriting model for Chinese patent abstracts based on reinforcement learning[J]. Data Analysis and Knowledge Discovery, 2021, 5(7): 59-69.
[25] HU B, CHEN Q, ZHU F. LCSTS: a large scale Chinese short text summarization dataset[J]. arXiv:1506.05865, 2015.
[26] GAO S, CHEN X, LI P, et al. Abstractive text summarization by incorporating reader comments[C]//Proceedings of the AAAI Conference on Artificial Intelligence, 2019: 6399-6406.
[27] WANG D, CHEN J, WU X, et al. CNewSum: a large-scale Chinese news summarization dataset with human-annotated adequacy and deducibility level[J]. arXiv: 2110.10874, 2021.
[28] LIN C Y. ROUGE: a package for automatic evaluation of summaries[C]//Proceedings of the Worlcshop on Text Summarization Branches Out, 2004: 74-81.
[29] LIU Y, LIU P, RADEV D, et al. BRIO: bringing order to abstractive summarization[J]. arXiv: 2203.16804, 2022.
[30] 张乐, 杜一凡, 吕学强, 等. STNLTP: 一种基于集成策略的中文专利摘要生成模型[J].数据分析与知识发现, 2022, 6(7): 107-117.
ZHANG L, DU Y F, LV X Q, et al. STNLTP: generating Chinese patent abstracts based on integrated strategy[J]. Data Analysis and Knowledge Discovery, 2022, 6(7): 107-117.