Using aspect-based sentiment analysis methods to analyze student evaluation texts is of significant importance for promoting the improvement of teaching quality. Existing aspect-based sentiment analysis methods include discriminative-based methods and generative-based methods. However, discriminative-based methods do not fully consider the associative semantics between aspect-level sentiment elements, while generative methods are overly dependent on template quality and still have room for improvement in understanding implicit sentiments. To address these issues, we propose a collaborative approach for aspect-level teaching evaluation texts based on large and small models. This model uses reasoning strategies based on large language models to deeply understand the semantics of evaluation texts and reason about implicit emotions, clarifying ambiguous emotions to enhance the model's ability to identify implicit emotions. Subsequently, we propose a dual-level interaction strategy based on small models, which fully captures contextual information through shallow-level and deep-level interactions, achieving fine-grained analysis of implicit emotions in teaching evaluation texts. The experimental results demonstrate that on the Laptop, Restaurant public datasets and self-constructed teaching evaluation text dataset, the F1 values of this model are 0.24, 0.60 and 1.05 percentage points higher than the optimal baseline model, respectively. A series of ablation and comparison experiments further confirm the method's effectiveness.
为了解决以上问题,本文提出了一种基于大小语言模型协同的评教文本方面级情感分析模型(Collaborative Approach Based on Large and Small Language Models,CO-LSM)。本文的主要贡献如下: 1) 针对现有方法对隐式情感的感知能力不足的问题,设计了基于大语言模型的隐式情感增强策略,对评价文本进行深层次的语义理解和隐式情感推理; 2) 针对现有方法未充分考虑方面级情感元素之间的关联语义的问题,设计了基于小模型的双层交互策略,通过浅层和深层交互充分捕捉评教文本的上下文信息; 3) 针对现有教学评价文本分析语义解析粒度不足的问题,提出大小语言模型协同的方面四元组抽取框架,以全面挖掘学生评教文本中蕴含的信息,及时为教学改进提供反馈意见,实现教师个性化服务,促进教育评价的科学化、规范化和智能化。
为了验证本文提出模型的有效性,本文在Laptop数据集、Restaurant数据集和自构建的教学评价文本数据集(Teaching Evaluation Text Dataset,TETD)上进行实验,数据分布如表1所示。其中,Restaurant为餐厅评论数据集,Laptop为笔记本电脑评论数据集,TETD为自构建的教学评价数据集。TETD数据集采用H大学的真实评教文本数据,根据李建龙等[29]设计的学生评教指标对教师教学方面进行方面类别划分,包含教学态度、教学内容、教学方法、教学效果和教学能力5项评价指标。并在标注过程中邀请两位数据标注专家进行双盲标注,以减少主观标注偏差,提高标注数据的质量。
为了验证所提出的大模型生成(Large Language Model Generation,LG)和深层次交互(Deep Level Interaction,DLI)的有效性,本文在三个数据集上进行了消融实验,结果如表3所示。实验结果表明,在引入大模型生成和深层次交互两个模块后,模型性能显著提升。这说明大模型生成和深层次交互两个模块的结合能够实现模型的协同增效,更有效地捕捉和分析文本中的隐式情感促进更深层次的语义分析和信息整合。
YEX L, QINY X, XIAL L. Logical reasoning, realistic dilemmas, and potential paths of evaluation of teachers teaching in universities in the perspective of scholarship of teaching[J]. Forum on Contemporary Education, 2024, 23(5): 33-40. DOI: 10.13694/j.cnki.ddjylt.20240628.002(Ch ).
ZHANGY C, LIM. Ideal and reality of college students’ evaluation of teaching: From the perspective of stakeholders[J]. Journal of China Examinations, 2022(9): 29-38. DOI: 10.19360/j.cnki.11-3303/g4.2022.09.005(Ch ).
ZHANGH X, PENGJ J. A deep semantic mining model based on aspect-level sentiment analysis[J]. Acta Electronica Sinica, 2024, 52(7): 2307-2319. DOI: 10.12263/DZXB.20230037(Ch ).
GUOX W, LAIH, YUZ T, et al. Emotion classification of case-related microblog comments integrating emotional knowledge[J]. Chinese Journal of Computers, 2021, 44(3): 564-578. DOI: 10.11897/SP.J.1016.2021.00564(Ch ).
[9]
TSENGC W, CHOUJ J, TSAIY C. Text mining analysis of teaching evaluation questionnaires for the selection of outstanding teaching faculty members[J]. IEEE Access, 2018, 6: 72870-72879. DOI: 10.1109/ACCESS.2018.2878478 .
KONGF Y, CHENG. Sentiment analysis of teaching evaluation text based on improved bidirectional LSTM[J]. Computer Engineering and Design, 2022, 43(12): 3580-3587. DOI: 10.16208/j.issn1000-7024.2022.12.035(Ch ).
PENGW Z, XIAJ L, WANQ Z, et al. Modeling aspect sentiment quad prediction as cloze task[J]. Chinese Journal of Computers, 2024, 47(8): 1744-1768. DOI: 10.11897/SP.J.1016.2024.01744(Ch ).
[14]
CAIH J, XIAR, YUJ F. Aspect-category-opinion-sentiment quadruple extraction with implicit aspects and opinions[C]//Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). Stroudsburg: Association for Computation Linguistics, 2021: 340-350. DOI:10.18653/v1/2021.acl-long.29 .
[15]
ZHANGW X, LIX, DENGY, et al. Towards generative aspect-based sentiment analysis[C]//Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 2: Short Papers). Stroudsburg: Association for Computation Linguistics, 2021: 504-510. DOI:10.18653/v1/2021.acl-short.64 .
[16]
MAOY, SHENY, YANGJ C, et al. Seq2Path: Generating sentiment tuples as paths of a tree[C]//Findings of the Association for Computational Linguistics. Stroudsburg: Association for Computation Linguistics, 2022: 2215-2225. DOI:10.18653/v1/2022.findings-acl.174 .
[17]
HUM T, WUY K, GAOH, et al. Improving aspect sentiment quad prediction via template-order data augmentation[C]//Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. Stroudsburg: Association for Computation Linguistics, 2022: 7889-7900. DOI:10.18653/v1/2022.emnlp-main.538 .
[18]
GOUZ B, GUOQ Y, YANGY J. MvP: Multi-view prompting improves aspect sentiment tuple prediction[C]//Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Stroudsburg: Association for Computation Linguistics, 2023: 4380-4397. DOI:10.18653/v1/2023.acl-long.240 .
LIANGY C, SONGS Q, ZHAOA J. Study on alienation of college students’ evaluation of teaching system[J]. The Party Building and Ideological Education in Schools, 2020(18): 55-57. DOI: 10.19865/j.cnki.xxdj.2020.18.018(Ch ).
ZHANGY C, LIM. Ideal and reality of college students’ evaluation of teaching:From the perspective of stakeholders[J]. Journal of China Examinations, 2022, 32(9):29-38. DOI:10.19360/j.cnki.11-3303/g4.2022.09.005(Ch ).
MAX L, ZHONGK D, LIUL C. Research of the validity about SETQ by big data analysis[J]. China Educational Technology, 2014(10): 78-84. DOI: 10.3969/j.issn.1006-9860.2014.10.013(Ch ).
[25]
LIG M, HOUG Y, WANGX J, et al. A multivariate generalizability theory approach to college students’ evaluation of teaching[J]. Frontiers in Psychology, 2018, 9(1): 1065. DOI:10.3389/fpsyg.2018.01065 .
[26]
LINQ K, ZHUY F, ZHANGS F, et al. Lexical based automated teaching evaluation via students’ short reviews[J]. Computer Applications in Engineering Education, 2019, 27(1): 194-205. DOI:10.1002/cae.22068 .
[27]
ONANA. Mining opinions from instructor evaluation reviews: A deep learning approach[J]. Computer Applications in Engineering Education, 2020, 28(1): 117-138. DOI:10.1002/cae.22179 .
WUQ L, DANGY G, XIONGS W, et al. Sentiment analysis model of students’ teaching evaluation text based on hybrid feature network[J]. Computer Engineering, 2023, 49(11): 24-29. DOI: 10.19678/j.issn.1000-3428.0066181(Ch ).
[30]
WANGW Y, PANS J, DAHLMEIERD, et al. Coupled multi-layer attentions for co-extraction of aspect and opinion terms[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2017, 31(1): 3316-3322. DOI:10.1609/aaai.v31i1.10974 .
[31]
CAIH J, TUY F, ZHOUX S, et al. Aspect-category based sentiment analysis with hierarchical graph convolutional network[C]//Proceedings of the 28th International Conference on Computational Linguistics. Stroudsburg: Association for Computation Linguistics,2020: 833-843. DOI:10.18653/v1/2020.coling-main.72 .
[32]
PENGH Y, XUL, BINGL D, et al. Knowing what, how and why: A near complete solution for aspect-based sentiment analysis[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(5): 8600-8607. DOI:10.1609/aaai.v34i05.6383 .
HUANGL D, HUH J, CHENJ Y, et al. Aspect-based sentiment analysis method based on knowledge enhancement and multi-layer attention mechanism[J]. Journal of Wuhan University(Natural Science Edition), 2024, 70(4):473-481. DOI: 10.14188/j.1671-8836.2023.0014(Ch ).
[35]
SUG X, WUM M, HUANGZ Q, et al. Refine, align, and aggregate: Multi-view linguistic features enhancement for aspect sentiment triplet extraction[C]//Findings of the Association for Computational Linguistics 2024. Stroudsburg: Association for Computation Linguistics, 2024: 3212-3228. DOI:10.18653/v1/2024.findings-acl.191 .
[36]
ZHANGW X, DENGY, LIX, et al. Aspect sentiment quad prediction as paraphrase generation[C]//Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. Stroudsburg: Association for Computation Linguistics, 2021: 9209-9219. DOI:10.18653/v1/2021.emnlp-main.726 .
[37]
WANGS, ZHOUJ, SUNC, et al. Causal intervention improves implicit sentiment analysis[C]//Proceedings of the 29th International Conference on Computational Linguistics. Stroudsburg: Association for Computation Linguistics, 2022:6966-6977.
[38]
CUIJ, FUKUMOTOF, WANGX F, SUZUKIY, et al. Aspect-category enhanced learning with a neural coherence model for implicit sentiment analysis[C]//Findings of the Association for Computational Linguistics EMNLP 2023. Stroudsburg: Association for Computation Linguistics, 2023: 11345-11358. DOI:10.18653/v1/2023.findings-emnlp.759 .
LIUY, ZENGZ M, SUNS Q. Research on aspect-level sentimen analysis based on LLM sentimen enhancement[J/OL]. Data Analysis and Knowledge Discovery, 2024: 1-17. (2024-12-10).
LIJ L, LIL G. Construction and application of university teaching quality evaluation index based on teaching evaluation text mining[J]. Chongqing Higher Education Research, 2024, 12(3): 11-24. DOI: 10.15998/j.cnki.issn1673-8012.2024.03.002(Ch ).
[43]
KIMJ, HEO R, SEOY, et al. Self-consistent reasoning-based aspect-sentiment quad prediction with extract-then-assign strategy[C]//Findings of the Association for Computational Linguistics ACL 2024. Stroudsburg: Association for Computational Linguistics, 2024: 7295-7303. DOI:10.18653/v1/2024.findings-acl.435 .