提示学习启发的无监督情感风格迁移研究

doi:10.3778/j.issn.1002-8331.2211-0317

摘要/Abstract

摘要： 文本样式迁移是在保留文本内容信息的同时移植具有某种所需样式属性的文本生成任务。为了提高在非平行样式语料下的迁移质量，提出了一种指导填充掩码模型将句子重写为目标样式的新方法。该方法总体上基于“删除-检索-生成”样式迁移框架，但采用大型无监督预训练语言模型和Transformer架构。根据Transformer的工作机理，改进了从源语句中筛选样式属性的方法，通过提示学习方法挖掘预训练模型的内部知识以实现对目标样式词的生成。在两个情感基准数据集上的实验表明，该方法在文本样式迁移任务上明显优于现有的编辑类方法，综合指标的相对分数平均提高了14%以上。

关键词: 文本生成, 文本样式迁移, 情感迁移, 提示学习, Transformer

Abstract: Text style transfer is the task of transferring text generation with certain desired style properties while preserving the original text content. In order to improve the transfer quality under non-parallel style corpus, this paper proposes a new method to guide the fill-mask model to rewrite the sentence into the target style. Overall, this approach is based on the delete-retrieve-generate style transfer framework, but employs a large unsupervised pre-trained language model and Transformer architecture. According to the working principle of Transformer, firstly, the method of filtering style attributes from the source sentence is improved, and then the internal knowledge of the pre-trained model is mined by the prompt learning method to generate the target style words. Experiments on two sentiment benchmark datasets show that the method outperforms existing editing methods, with an average improvement of more than 14% in relative scores on the comprehensive metrics.

Key words: text generation, text style transfer, sentiment transfer, prompt learning, Transformer

蔡国永, 李安庆. 提示学习启发的无监督情感风格迁移研究[J]. 计算机工程与应用, 2024, 60(5): 146-155.

CAI Guoyong, LI Anqing. Prompt-Learning Inspired Approach to Unsupervised Sentiment Style Transfer[J]. Computer Engineering and Applications, 2024, 60(5): 146-155.

参考文献

[1] XU W, RITTER A, DOLAN B, et al. Paraphrasing for style[C]//Proceedings of the COLING 2012, 2012: 2899-2914.
[2] ZHOU G, LUO P, CAO R, et al. Mechanism aware neural machine for dialogue response generation[C]//Proceedings of the 31st AAAI Conference on Artificial Intelligence, 2017.
[3] REDDY S, KNIGHT K. Obfuscating gender in social media writing[C]//Proceedings of the First Workshop on NLP and Computational Social Science, 2016: 17-26.
[4] RAO S, TETREAULT J. Dear sir or madam, may I introduce the GYAFC dataset: corpus, benchmarks and metrics for formality style transfer[C]//Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologie, 2018: 129-140.
[5] YANG C, SUN M, YI X, et al. Stylistic Chinese poetry generation via unsupervised style disentanglement[C]//Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, 2018: 3960-3969.
[6] MADAAN A, SETLUR T, PAREKH B, et al. Politeness transfer: a tag and generate approach[J]. arXiv:2004.14257,2020.
[7] FU Z, TAN X, PENG N, et al. Style transfer in text: exploration and evaluation[C]//Proceedings of the AAAI Conference on Artificial Intelligence, 2018.
[8] 胡盛伟, 李弼程, 林孔杰，等. MaskAE: 基于无监督的短文本情感迁移方法[J]. 中文信息学报, 2020, 34(2): 105-112.
HU S W, LI B C, LIN K J, XIONG Y, et al. MaskAE: unsupervised short text sentiment transfer method[J]. Journal of Chinese Information Processing, 2020, 34(2): 105-112.
[9] HE J, WANG X, NEUBIG G, et al. A probabilistic formulation of unsupervised text style transfer[C]//Proceedings of the International Conference on Learning Representations (ICLR), 2020.
[10] SHEN T, LEI T, BARZILAY R, et al. Style transfer from non-parallel text by cross-alignment[C]//Advances in Neural Information Processing Systems, 2017.
[11] LUO F, LI P, ZHOU J, et al. A dual reinforcement learning framework for unsupervised text style transfer[J]. arXiv:1905.10060,2019.
[12] DAI N, LIANG J, QIU X, et al. Style transformer: unpaired text style transfer without disentangled latent representation[C]//Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, 2019: 5997-6007.
[13] ZHANG Z, REN S, LIU S, et al. Style transfer as unsupervised machine translation[J]. arXiv:1808.07894,2018.
[14] SHI Y, ZHANG S, ZHOU C, et al. Gtae: graph transformer-based auto-encoders for linguistic-constrained text style transfer[J]. ACM Transactions on Intelligent Systems and Technology (TIST), 2021, 12(3): 1-16.
[15] LI J, JIA R, HE H, et al. Delete, retrieve, generate: a simple approach to sentiment and style transfer[C]//Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2018: 1865-1874.
[16] WU X, ZHANG T, ZANG L, et al. Mask and infill: applying masked language model to sentiment transfer[C]//Proceedings of the International Joint Conference on Artificial Intelligence, 2019.
[17] MALMI E, SEVERYN A, ROTHE S. Unsupervised text style transfer with padded masked language models[C]//Proceedings of the Conference on Empirical Methods in Natural Language Processing, 2020: 8671-8680.
[18] SUDHAKAR A, UPADHYAY B, MAHESWARAN A. Transforming delete, retrieve, generate approach for controlled text style transfer[C]//Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, 2019: 3260-3270.
[19] SCAO T L, RUSH A M. How many data points is a prompt worth? [C]//Proceedings of the North American Chapter of the Association for Computational Linguistics, 2021.
[20] MANNING C D, SURDEANU M, BAUER J, et al. The Stanford CoreNLP natural language processing toolkit[C]//Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics: System Demonstrations, 2014: 55-60.
[21] KEENAN, WILLIAM J F. Sacre Bleu: faith, fashion and freedom: marist foundation garments 1817-1862[M]. London;New York: Routledge, 2017: 132-153.
[22] HEAFIELD K. KenLM: faster and smaller language model queries[C]//Proceedings of the Sixth Workshop on Statistical Machine Translation, 2011: 187-197.