文本情感分析是自然语言处理领域的关键研究方向之一，它在产品反馈分析、公共舆论监控、个性化推荐服务等多个应用场景中发挥着重要作用。随着社交媒体的迅速渗透，网络上的用户评论数据量呈现出爆炸性增长，这些评论中蕴含的丰富情感信息对于消费者了解产品特性、企业优化产品与服务、以及政府机构提升政务服务质量具有不可估量的价值。因此，深入挖掘在线媒体评论中的情感信息，对于学术研究和实际应用均具有极其重要的意义。然而现有的文本情感分析模型仍然存在着语义获取不充分等问题，同时面对复杂多变的网络信息，大量隐式情感表达给文本情感分析模型在性能表现上带来了挑战。鉴于这些问题，本研究选择文本评论内容作为研究对象，对不同类型的评论文本进行了细致的分析和研究，旨在提高情感分析模型的准确性。

（1）针对现有的情感分类模型在局部建模时邻域标记不恰当，以及在上下文建模时没有充分表征上下文信息等问题，本文提出了一种基于BERT的融合局部和上下文信息的方面情感分析模型（A Local and Context Fusion Sentiment Analysis Model Based on BERT, LCA-BERT）。具体地，使用动态掩码矩阵替代BERT模型中自注意力网络中的静态掩码矩阵，使模型能够更有效地获取局部信息；其次，引入了“准”注意力计算和深度全局上下文的方法，这种方法不仅降低了文本噪声对模型的影响，而且充分获取了上下文语义信息。经过与基线模型的对比实验表明，本文模型在方面情感分析任务上的准确率和AUC分别达到了94.5%和97.8%，验证了本文模型的先进性。

（2）针对情感语句中缺乏明确情感特征，模型无法准确分析情感极性等问题，本文提出了一种融合大语言模型的三级联合提示隐式情感分析方法（A Three-Level Joint Prompt-tuning Sentiment Analysis Method Incorporating LLMs, TPISA），将大语言模型与本地预训练模型相结合，使用多级推理的方式逐级得出目标的方面、潜在观点，使模型能够更轻松地推理出最终的情感极性。前两级提示利用大型语言模型丰富的世界知识，丰富了情感语句的情感信息；然后，将前两级提示得到的方面和潜在意见与上下文连接起来，作为第三级提示的输入，使预训练的模型能够从标签词汇中获得丰富的语义知识，增强模型的学习能力。实验证明，本文提出的模型在SemEval14 Laptop和Restaurant数据集上比基线模型的性能表现更好，准确率分别达到了77.93%和82.31%，证明了本文模型在隐式情感分析任务上的先进性，为情感分析领域的后续研究提供了有益的参考。

Textual sentiment analysis, as a pivotal research direction within the realm of natural language processing, has been extensively applied across various application scenarios, including product analysis, public sentiment monitoring, and personalized services. With the rapid proliferation of social media, there has been a marked increase in user-generated comment data on online platforms. The affective content embedded within these textual reviews is instrumental in aiding users in their understanding of products, facilitating enterprises and platforms in refining product quality, and enabling governments to enhance the caliber of public services. Consequently, the excavation of affective information from online media reviews holds significant research implications. However, prevailing models of textual sentiment analysis are not without their shortcomings, as they often fall short in the comprehensive acquisition of semantics. Moreover, the intricate and mutable nature of online information presents a formidable challenge, with a plethora of implicit emotional expressions adversely affecting the performance of sentiment analysis models. In light of these issues, the present study focuses on online textual reviews as its subject of investigation, delving into different types of review texts to address these challenges and refine the analytical models accordingly.

(1) The current sentiment classification models grapple with the challenge of inadequate neighborhood labeling in local modeling and a shortfall in capturing the full spectrum of contextual information during contextual modeling. In response to these shortcomings, this paper introduces an innovative sentiment analysis model that harnesses the power of BERT for the fusion of local and contextual semantic information, termed A Local and Context Fusion Sentiment Analysis Model Based on BERT (LCA-BERT). Specifically, this study replaces the static mask matrix in BERT's self-attention network with a dynamic mask matrix, enabling the model to more effectively capture local details. Furthermore, the introduction of "quasi" attention weights and a method for deep global context reduces the impact of textual noise on the model while fully acquiring contextual semantic information. Comparative experiments with baseline models demonstrate that the proposed model achieves an accuracy of 94.5% and an AUC of 97.8% in aspect-based sentiment analysis tasks, thereby validating the model's sophistication and efficacy.

(2) In response to the deficiency of explicit sentiment lexicon within sentiment expressions and the models' consequent struggle to accurately discern sentiment polarity, this paper proposes a Three-Level Joint Prompt-tuning Sentiment Analysis Method Incorporating Large Language Models (TPISA). This method integrates a large language models(LLMs) with a locally pre-trained model, employing a multi-level inference approach to progressively deduce the target aspect and latent opinion, thereby enabling the model to more readily infer the ultimate sentiment polarity. The first two levels of prompt leverage the extensive world knowledge of large language models to enrich the emotional information within sentiment phrases. Subsequently, the aspects and latent opinions derived from the first two levels prompt are connected with the context, serving as input for the third level prompt, enabling the pre-trained model to acquire rich semantic knowledge from the labeled vocabulary and enhance its learning capabilities. Experiments have confirmed that the model proposed in this paper outperforms baseline models on the SemEval-2014 Laptop and Restaurant datasets, with accuracy rates of 77.93% and 82.31%, respectively. This proves the progressiveness of the model in implicit emotion analysis, and offers valuable insights for research in the field of implicit sentiment analysis.

[1] CHATTERJEE A, NARAHARI K N, JOSHI M, et al. SemEval-2019 task 3: EmoContext contextual emotion detection in text[C/OL]//NAACL HLT 2019 - International Workshop on Semantic Evaluation, SemEval 2019, Proceedings of the 13th Workshop. 2019.

[2] SAILUNAZ K, DHALIWAL M, ROKNE J, et al. Emotion detection from text and speech: a survey[J/OL]. Social Network Analysis and Mining, 2018, 8(1).

[3] DAVE CHAFFEY. Global social media statistics research summary 2022[EB]//Smart Insights. 2023.

[4] KANG X, REN F, WU Y. Exploring latent semantic information for textual emotion recognition in blog articles[J/OL]. IEEE/CAA Journal of Automatica Sinica, 2018, 5(1): 204-216.

[5] SHI W, LI F, LI J, et al. Effective token graph modeling using a novel labeling strategy for structured sentiment analysis[J]. arXiv preprint arXiv:2203.10796, 2022.

[6] ZHU L, XU M, BAO Y, et al. Deep learning for aspect-based sentiment analysis: a review[J/OL]. PeerJ Computer Science, 2022, 8.

[7] RAGHUNATHAN N, SARAVANAKUMAR K. Challenges and Issues in Sentiment Analysis: A Comprehensive Survey[J/OL]. IEEE Access, 2023, 11: 69626-69642.

[8] 周咏梅, 杨佳能, 阳爱民. 面向文本情感分析的中文情感词典构建方法[J]. 《山东大学学报（工学版）》, 2013: 27-33.

[9] HUTTO C J, GILBERT E. VADER: A Parsimonious Rule-based Model for Sentiment Analysis of Social Media Text[R/OL]. (2014).

[10] 陈强, 何炎祥. 基于句法分析的跨语言情感分析[J]. 北京大学学报(自然科学版), 2014, 50(1).

[11] LIN C Y, HOVY E. Automatic evaluation of summaries using N-gram co-occurrence statistics[C/OL]//Proceedings of the 2003 Human Language Technology Conference of the North American Chapter of the Association for Computational Linguistics, HLT-NAACL 2003. 2003.

[12] 解军, 邢进生. 基于KNN算法的新浪微博用户行为分析及预测[J]. 《山西师范大学学报（自然科学版）》, 2016: 38-45.

[13] BRAIG N, BENZ A, VOTH S, et al. Machine Learning Techniques for Sentiment Analysis of COVID-19-Related Twitter Data[M/OL]//IEEE Access. (2023).

[14] MULLEN T, COLLIER N. Sentiment analysis using support vector machines with diverse information sources[C]//Proceedings of the 2004 Conference on Empirical Methods in Natural Language Processing, EMNLP 2004 - A meeting of SIGDAT, a Special Interest Group of the ACL held in conjunction with ACL 2004. 2004.

[15] PAK A, PAROUBEK P. LNCS 6637 - Text Representation Using Dependency Tree Subgraphs for Sentiment Analysis[R].

[16] JOHN lafferty, ANDREW M, FERNANDO C N P. Conditional Random Fields: Probabilistic Models for Segmenting and Labeling Sequence Data[J/OL]. ICML ’01: Proceedings of the Eighteenth International Conference on Machine Learning, 2001, 2001(June).

[17] VALDIVIA A, LUZÓN M V, CAMBRIA E, et al. Consensus vote models for detecting and filtering neutrality in sentiment analysis[J/OL]. Information Fusion, 2018, 44.

[18] LE T. A hybrid method for text-based sentiment analysis[C/OL]//Proceedings - 6th Annual Conference on Computational Science and Computational Intelligence, CSCI 2019. 2019.

[19] DETTMERS T, MINERVINI P, STENETORP P, et al. Convolutional 2D knowledge graph embeddings[C/OL]//32nd AAAI Conference on Artificial Intelligence, AAAI 2018. 2018.

[20] KHATUA A, KHATUA A, CAMBRIA E. Predicting political sentiments of voters from Twitter in multi-party contexts[J/OL]. Applied Soft Computing Journal, 2020, 97.

[21] YIN R, LI P, WANG B. Sentiment Lexical-Augmented Convolutional Neural Networks for Sentiment Analysis[C/OL]//Proceedings - 2017 IEEE 2nd International Conference on Data Science in Cyberspace, DSC 2017. 2017.

[22] CONNEAU A, SCHWENK H, CUN Y Le, et al. Very deep convolutional networks for text classification[C/OL]//15th Conference of the European Chapter of the Association for Computational Linguistics, EACL 2017 - Proceedings of Conference: Volume 2. 2017.

[23] J A K, TRUEMAN T E, CAMBRIA E. A Convolutional Stacked Bidirectional LSTM with a Multiplicative Attention Mechanism for Aspect Category and Sentiment Detection[J/OL]. Cognitive Computation, 2021, 13(6): 1423-1432.

[24] LIANG B, SU H, GUI L, et al. Aspect-based sentiment analysis via affective knowledge enhanced graph convolutional networks[J/OL]. Knowledge-Based Systems, 2022, 235.

[25] DO H H, PRASAD P W C, MAAG A, et al. Deep Learning for Aspect-Based Sentiment Analysis: A Comparative Review[M/OL]//Expert Systems with Applications. (2019).

[26] LIU H, CHATTERJEE I, ZHOU M, et al. Aspect-Based Sentiment Analysis: A Survey of Deep Learning Methods[M/OL]//IEEE Transactions on Computational Social Systems. (2020).

[27] ZHANG W, LI X, DENG Y, et al. A Survey on Aspect-Based Sentiment Analysis: Tasks, Methods, and Challenges[J/OL]. IEEE Transactions on Knowledge and Data Engineering, 2023, 35(11).

[28] LI X, BING L, ZHANG W, et al. Exploiting bert for end-to-end aspect-based sentiment analysis_[C/OL]//W-NUT@EMNLP 2019 - 5th Workshop on Noisy User-Generated Text, Proceedings. 2019.

[29] SUN C, HUANG L, QIU X. Utilizing BERT for aspect-based sentiment analysis via constructing auxiliary sentence[C]//NAACL HLT 2019 - 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies - Proceedings of the Conference: Volume 1. 2019.

[30] YANG B, LI J, WONG D F, et al. Context-aware self-attention networks[C/OL]//33rd AAAI Conference on Artificial Intelligence, AAAI 2019, 31st Innovative Applications of Artificial Intelligence Conference, IAAI 2019 and the 9th AAAI Symposium on Educational Advances in Artificial Intelligence, EAAI 2019. 2019.

[31] SONG Y, WANG J, JIANG T, et al. Attentional Encoder Network for Targeted Sentiment Classification[J/OL]. 2019.

[32] RIETZLER A, STABINGER S, OPITZ P, et al. Adapt or Get Left Behind: Domain Adaptation through BERT Language Model Finetuning for Aspect-Target Sentiment Classification[J/OL]. 2019.

[33] WU Z, ONG D C. Context-Guided BERT for Targeted Aspect-Based Sentiment Analysis[J/OL]. Proceedings of the AAAI Conference on Artificial Intelligence, 2021, 35(16): 14094-14102.

[34] CHEN Y, ZHUANG T, GUO K. Memory network with hierarchical multi-head attention for aspect-based sentiment analysis[J/OL]. Applied Intelligence, 2021, 51(7).

[35] GU T, ZHAO H, LI M. Effective inter-aspect words modeling for aspect-based sentiment analysis[J/OL]. Applied Intelligence, 2023, 53(4): 4366-4379.

[36] YANG H, LI K. Improving Implicit Sentiment Learning via Local Sentiment Aggregation[J/OL]. 2021.

[37] LIAO J, WANG S, LI D. Identification of fact-implied implicit sentiment based on multi-level semantic fused representation[J]. Knowledge-Based Systems, 2019, 165(FEB.1): 197-207.

[38] BALAHUR A, HERMIDA J M, MONTOYO A. Detecting implicit expressions of emotion in text: A comparative analysis[J]. Decision Support Systems, 2012, 53(4): 742-753.

[39] SHUTOVA E, TEUFEL S, KORHONEN A. Statistical metaphor processing[J/OL]. Computational Linguistics, 2013, 39(2).

[40] DENG L, WIEBE J. Sentiment propagation via implicature constraints[C/OL]//14th Conference of the European Chapter of the Association for Computational Linguistics 2014, EACL 2014. 2014.

[41] XIANG C, REN Y, JI D. Identifying Implicit Polarity of Events by Using an Attention-Based Neural Network Model[J/OL]. IEEE Access, 2019, 7.

[42] WEI J, WANG X, SCHUURMANS D, et al. Chain-of-thought prompting elicits reasoning in large language models[J]. Advances in Neural Information Processing Systems, 2022, 35: 24824-24837.

[43] LI Z, ZOU Y, ZHANG C, et al. Learning Implicit Sentiment in Aspect-based Sentiment Analysis with Supervised Contrastive Pre-Training[C/OL]//EMNLP 2021 - 2021 Conference on Empirical Methods in Natural Language Processing, Proceedings. 2021.

[44] YANG S, XING L, LI Y, et al. Implicit sentiment analysis based on graph attention neural network[J/OL]. Engineering Reports, 2022, 4(1).

[45] MIKOLOV T, CHEN K, CORRADO G, et al. Efficient estimation of word representations in vector space[C]//1st International Conference on Learning Representations, ICLR 2013 - Workshop Track Proceedings. 2013.

[46] PENNINGTON J, SOCHER R, MANNING C D. GloVe: Global vectors for word representation[C/OL]//EMNLP 2014 - 2014 Conference on Empirical Methods in Natural Language Processing, Proceedings of the Conference. 2014.

[47] DEVLIN J, CHANG M W, LEE K, et al. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding[M]. 2019.

[48] LIU P, YUAN W, FU J, et al. Pre-train, Prompt, and Predict: A Systematic Survey of Prompting Methods in Natural Language Processing[J/OL]. ACM Computing Surveys, 2023, 55(9).

[49] GUO M, ZHANG Y, LIU T. Gaussian Transformer: A Lightweight Approach for Natural Language Inference[J/OL]. Proceedings of the AAAI Conference on Artificial Intelligence, 2019, 33(01): 6489-6496.

[50] FAN Z, GONG Y, LIU D, et al. Mask Attention Networks: Rethinking and Strengthen Transformer[C/OL]//Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. Stroudsburg, PA, USA: Association for Computational Linguistics, 2021: 1692-1701.

[51] TAY Y, TUAN L A, ZHANG A, et al. Compositional de-attention networks[C]//Advances in Neural Information Processing Systems: Volume 32. 2019.

[52] PETERS M E, NEUMANN M, IYYER M, et al. Deep Contextualized Word Representations[C/OL]//WALKER M, JI H, STENT A. Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers). New Orleans, Louisiana: Association for Computational Linguistics, 2018: 2227-2237.

[53] VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[J]. Advances in neural information processing systems, 2017, 30.

[54] BRITZ D, GOLDIE A, LUONG M T, et al. Massive exploration of neural machine translation architectures[C/OL]//EMNLP 2017 - Conference on Empirical Methods in Natural Language Processing, Proceedings. 2017.

[55] HOCHREITER S, SCHMIDHUBER J. Long Short-Term Memory[J/OL]. Neural Computation, 1997, 9(8): 1735-1780.

[56] LIU F, COHN T, BALDWIN T. Recurrent entity networks with delayed memory update for targeted aspect-based sentiment analysis[C/OL]//NAACL HLT 2018 - 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies - Proceedings of the Conference: Volume 2. 2018.

[57] PHAN M H, OGUNBONA P. Modelling context and syntactical features for aspect-based sentiment analysis[C/OL]//Proceedings of the Annual Meeting of the Association for Computational Linguistics. 2020.

[58] 黄山成, 韩东红, 乔百友, 等. 基于ERNIE2.0-BiLSTM-Attention的隐式情感分析方法[J]. 小型微型计算机系统, 2021, 42(12): 12-2485.

[59] YE J, CHEN X, XU N, et al. A Comprehensive Capability Analysis of GPT-3 and GPT-3.5 Series Models[J]. CoRR, 2023, abs/2303.10420.

[60] LIU J, LIU A, LU X, et al. Generated Knowledge Prompting for Commonsense Reasoning[C]//Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2022: 3154-3169.

[61] 张心月, 刘蓉*, 魏驰宇, 方可. 融合提示知识的方面级情感分析方法[J]. 计算机应用, 2023, 43(9): 2753-2759.

[62] ZHANG C, LI Q, SONG D. Aspect-based sentiment classification with aspect-specific graph convolutional networks[C/OL]//EMNLP-IJCNLP 2019 - 2019 Conference on Empirical Methods in Natural Language Processing and 9th International Joint Conference on Natural Language Processing, Proceedings of the Conference. 2019. DOI:10.18653/v1/d19-1464.

[63] BENGIO Y, OTHERS. From system 1 deep learning to system 2 deep learning[C]//Neural Information Processing Systems. 2019.

[64] PONTIKI M, GALANIS D, PAVLOPOULOS J, 等. SemEval-2014 Task 4: Aspect Based Sentiment Analysis[C/OL]//8th International Workshop on Semantic Evaluation, SemEval 2014 - co-located with the 25th International Conference on Computational Linguistics, COLING 2014, Proceedings. 2014. DOI:10.3115/v1/s14-2004.

[65] YANG Z, DAI Z, YANG Y, et al. Xlnet: Generalized autoregressive pretraining for language understanding[J]. Advances in neural information processing systems, 2019, 32.

[66] LAN Z, CHEN M, GOODMAN S, et al. ALBERT: A LITE BERT FOR SELF-SUPERVISED LEARNING OF LANGUAGE REPRESENTATIONS[C]//8th International Conference on Learning Representations, ICLR 2020. 2020.

[67] SCHICK T, SCHÜTZE H. Exploiting cloze questions for few shot text classification and natural language inference[C/OL]//EACL 2021 - 16th Conference of the European Chapter of the Association for Computational Linguistics, Proceedings of the Conference. 2021.

[68] HU S, DING N, WANG H, et al. Knowledgeable Prompt-tuning: Incorporating Knowledge into Prompt Verbalizer for Text Classification[M]. 2022.

[69] FAN F, FENG Y, ZHAO D. Multi-grained attention network for aspect-level sentiment classification[C/OL]//Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, EMNLP 2018. 2018.

[70] XU H, LIU B, SHU L, et al. BERT post-training for review reading comprehension and aspect-based sentiment analysis[C]//NAACL HLT 2019 - 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies - Proceedings of the Conference: Volume 1. 2019.

[71] JIANG Q, CHEN L, XU R, et al. A challenge dataset and effective models for aspect-based sentiment analysis[C/OL]//EMNLP-IJCNLP 2019 - 2019 Conference on Empirical Methods in Natural Language Processing and 9th International Joint Conference on Natural Language Processing, Proceedings of the Conference. 2019.

[72] 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/OL]. 山西大学学报（自然科学版）, 2023, 46(3): 509-517.