基于PPLM模板增强的零样本方面类别情感分析模型

doi:10.3778/j.issn.1673-9418.2406005

摘要/Abstract

摘要： 如今方面类别情感分析（ACSA）因标注数据稀缺而受限，如何在无特定情感标注数据下实现有效分析成为研究挑战。将零样本方面类别情感分析任务转换为自然语言推理（NLI）任务，针对传统提示模板面临语义表达不充分的问题，提出了一种基于PPLM文本限制生成模型的原因补充提示模板，结合情感极性和其原因，使得模板可以帮助模型更好地理解情感背后的原因和动机，从而提高情感分析的准确性和可解释性。为了进一步提升ACSA的分类性能，引入了性能逆序系数来确定文中多种提示模板的集成权重。在公共数据集MAMS和Restaurant的实验结果表明该模型相较于其他零样本ACSA模型，在准确率ACC上提升约7个百分点；PPLM原因补充提示模板可以提升NLI模型的零样本分类性能，其相较于其他较好传统模板在MF1上提升2.6个百分点；同时改进的权重确定策略也对模型在零样本情境下的情感分析能力有一定提升作用。

关键词: 方面类别情感分析, 零样本学习, 提示模板, PPLM文本限制生成

Abstract: Aspect category sentiment analysis (ACSA) is currently constrained by the scarcity of annotated data, making it a research challenge to achieve effective analysis without specific sentiment-labeled data. This paper transforms the zero-shot ACSA task into a natural language inference (NLI) task. Addressing the issue of inadequate semantic expression in traditional prompt templates, this paper proposes a cause supplementation prompt template based on the plug and play language model (PPLM) for text restriction generation. By combining sentiment polarity with its causes, the template helps the model better understand the reasons and motivations behind the sentiments, thereby improving the accuracy and interpretability of sentiment analysis. Furthermore, to enhance the classification performance of ACSA, the performance inversion coefficient is introduced to determine the ensemble weights of various prompt templates in the paper. Experimental results on the public datasets MAMS and Restaurant demonstrate that the proposed model outperforms other zero-shot ACSA models by approximately 7 percentage points in accuracy. The PPLM cause supplementation prompt template can enhance the zero-shot classification performance of NLI models, showing a 2.6 percentage points improvement in Macro F1 score compared with other better traditional templates. Additionally, the improved weight determination strategy also contributes to the model??s sentiment analysis capability in zero-shot scenarios.

Key words: aspect category sentiment analysis, zero-shot learning, prompt template, PPLM text-conditioned generation

凤丽洲, 李梦莎, 王友卫, 杨贵军. 基于PPLM模板增强的零样本方面类别情感分析模型[J]. 计算机科学与探索, 2025, 19(5): 1322-1333.

FENG Lizhou, LI Mengsha, WANG Youwei, YANG Guijun. Zero-Shot Aspect Category Sentiment Analysis Model with Enhanced PPLM Template[J]. Journal of Frontiers of Computer Science and Technology, 2025, 19(5): 1322-1333.

参考文献

[1] 张鲁宁, 左信, 刘建伟. 零样本学习研究进展[J]. 自动化学报, 2020, 46(1): 1-23.
ZHANG L N, ZUO X, LIU J W. Research and development on zero-shot learning[J]. Acta Automatica Sinica, 2020, 46(1): 1-23.
[2] HU M Q, LIU B, HU M Q, et al. Mining and summarizing customer reviews[C]//Proceedings of the 10th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York: ACM, 2004: 168-177.
[3] WANG Y Q, HUANG M L, ZHU X Y, et al. Attention-based LSTM for aspect-level sentiment classification[C]//Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing. Stroudsburg: ACL, 2016: 606-615.
[4] SUN C, HUANG L Y, QIU X P. Utilizing BERT for aspect-based sentiment analysis via constructing auxiliary sentence[C]//Proceedings of the Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. Stroudsburg: ACL, 2019: 380-385.
[5] SEOH R, BIRLE I, TAK M, et al. Open aspect target sentiment classification with natural language prompts[C]//Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. Stroudsburg: ACL, 2021: 6311-6322.
[6] 肖泽管, 陈清亮. 融合多种类型语法信息的属性级情感分析模型[J]. 计算机科学与探索, 2022, 16(2): 395-402.
XIAO Z G, CHEN Q L. Aspect-based sentiment analysis model with multiple grammatical information[J]. Journal of Frontiers of Computer Science and Technology, 2022, 16(2): 395-402.
[7] TIAN Y H, CHEN G M, SONG Y. Enhancing aspect-level sentiment analysis with word dependencies[C]//Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume. Stroudsburg: ACL, 2021: 3726-3739.
[8] LI W, ZHANG H, WANG M. Dialogue sentiment quadruple analysis: a multilingual dataset[C]//Proceedings of the 2023 International Conference on Computational Linguistics, 2023: 45-60.
[9] LING Y, YU J F, XIA R. Vision-language pre-training for multimodal aspect-based sentiment analysis[C]//Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics. Stroudsburg: ACL, 2022: 2149-2159.
[10] SOCHER R, GANJOO M, MANNING C D, et al. Zero-shot learning through cross-modal transfer[C]//Proceedings of the 27th International Conference on Neural Information Processing Systems, 2013: 935-943.
[11] SAPPADLA P V, NAM J, MENCIA E L, et al. Using semantic similarity for multi-label zero-shot classification of text documents[C]//Proceedings of the 23rd European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning, 2016: 423-428.
[12] RIOS A, KAVULURU R. Few-shot and zero-shot multi-label learning for structured label spaces[C]//Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing. Stroudsburg: ACL, 2018: 3132-3142.
[13] BROWN T B, MANN B, RYDER N, et al. Language models are few-shot learners[C]//Proceedings of the 34th International Conference on Neural Information Processing Systems. Cambridge: MIT Press, 2020: 1877-1901.
[14] HALDER K, AKBIK A, KRAPAC J, et al. Task-aware representation of sentences for generic text classification[C]//Proceedings of the 28th International Conference on Computational Linguistics. Stroudsburg: ACL, 2020: 3202-3213.
[15] WANG Y S, CHI T C, ZHANG R H, et al. PESCO: prompt-enhanced self contrastive learning for zero-shot text classification[C]//Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics. Stroudsburg: ACL, 2023: 14897-14911.
[16] ZHANG H, LI Y Z, ZHU T F, et al. Commonsense-based adversarial learning framework for zero-shot stance detection[J]. Neurocomputing, 2024, 563: 126943.
[17] YIN W P, HAY J, ROTH D. Benchmarking zero-shot text classification: datasets, evaluation and entailment approach[C]//Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing. Stroudsburg: ACL, 2019: 3914-3923.
[18] LIU J, TENG Z Y, CUI L Y, et al. Solving aspect category sentiment analysis as a text generation task[C]//Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. Stroudsburg: ACL, 2021: 4406-4416.
[19] ZHANG H X, ZHANG X F, HUANG H B, et al. Prompt-based meta-learning for few-shot text classification[C]//Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. Stroudsburg: ACL, 2022: 1342- 1357.
[20] PETRONI F, ROCKTASCHEL T, RIEDEL S, et al. Language models as knowledge bases?[C]//Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing. Stroudsburg: ACL, 2019: 2463-2473.
[21] SCHICK T, SCHVTZE H. Exploiting cloze-questions for few-shot text classification and natural language inference[C]//Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume. Stroudsburg: ACL, 2021: 255-269.
[22] SCHICK T, SCHVTZE H. It??s not just size that matters: small language models are also few-shot learners[C]//Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. Stroudsburg: ACL, 2021: 2339-2352.
[23] 王昱婷, 刘一伊, 张儒清, 等. 基于提示学习的文本隐式情感分类[J]. 山西大学学报(自然科学版), 2023, 46(3): 509-517.
WANG Y T, LIU Y Y, ZHANG R Q, et al. Learning implicit sentiment via prompt tuning[J]. Journal of Shanxi University (Natural Science Edition), 2023, 46(3): 509-517.
[24] JIANG Z B, XU F, ARAKI J, et al. How can we know what language models know?[J]. Transactions of the Association for Computational Linguistics, 2020, 8: 423-438.
[25] HAVIV A, BERANT J, GLOBERSON A. BERTese: learning to speak to BERT[C]//Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume. Stroudsburg: ACL, 2021: 3618-3623.
[26] GAO T Y, FISCH A, CHEN D Q. Making pre-trained language models better few-shot learners[C]//Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing. Stroudsburg: ACL, 2021: 3816-3830.
[27] LI J Y, TANG T Y, NIE J Y, et al. Learning to transfer prompts for text generation[C]//Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. Stroudsburg: ACL, 2022: 3506-3518.
[28] DIAO S Z, WANG P C, LIN Y, et al. Active prompting with chain-of-thought for large language models[EB/OL]. [2024-03-17]. https://arxiv.org/abs/2302.12246.
[29] LI Z K, PENG B L, HE P C, et al. Guiding large language models via directional stimulus prompting[EB/OL]. [2024-03-17]. https://arxiv.org/abs/2302.11520.
[30] FLOR M P, MARIA M, KLINGER R. Natural language inference prompts for zero-shot emotion classification in text across corpora[C]//Proceedings of the 29th International Conference on Computational Linguistics, 2022: 3506-3518.
[31] DATHATHRI S, MADOTTO A, LAN J, et al. Plug and play language models: a simple approach to controlled text generation[EB/OL]. [2024-03-17]. https://arxiv.org/abs/1912.02164.