融合字符与词语特征的混合神经网络情感分析模型

doi:10.3778/j.issn.1002-8331.2404-0236

摘要/Abstract

摘要： 汉语语句中没有明显的分隔符，这导致传统基于词语划分的编码模型可能会丢失句子内部的语义信息，特别是在处理情感分析等任务时显得更为复杂。为克服这一难题，借鉴字符级和词语级特征融合的思路，提出了一种融合特征模型。该模型将句子划分为两种级别的编码，采用Bi-GRU结构提取字符序列中的包含上下文信息的特征关系，并引入注意力机制，使用CNN网络结构提取词语之间的局部特征关系，利用不同大小的卷积核获得不同距离的局部特征，最后将二者特征进行融合，获得全局特征信息。在三个公开数据集Weibo、CIN和Chnsenticorp上的准确率分别达到了81.32%、76.03%和96.28%，相比于以字符编码为基础的MCNN-IFGS模型，分别提高了1.02个百分点、0.13个百分点和1.05个百分点，结果表明在中文情感分析任务中，融合特征模型的表现明显优于单独使用字符级或词语级特征的模型，能够显著提升模型的性能和鲁棒性，更有效地提取文本的语义信息。

关键词: 情感分析, 混合神经网络, 字符特征, 词语特征, 双向门控循环单元

Abstract: Chinese sentences lack explicit delimiters, making traditional word-based encoding models prone to losing internal semantic information—especially in sentiment analysis tasks. To address this, a feature fusion model is proposed that integrates character-level and word-level representations. It uses a Bi-GRU with attention to capture contextual features from character sequences and a CNN with multiple kernel sizes to extract local word-level features. The fused features yield comprehensive global representations. Tested on three public datasets—Weibo, CIN, and Chnsenticorp, the model achieves accuracies of 81.32%, 76.03%, and 96.28%, respectively, outperforming the character-based MCNN-IFGS model by 1.02, 0.13, and 1.05 percentage points. These results show that the fusion approach significantly enhances performance and robustness in Chinese sentiment analysis by effectively capturing semantic information.

Key words: sentiment analysis, hybrid neural networks, character features, word features, bidirectional gated recurrent units

李嘉琦, 杨环, 高辉. 融合字符与词语特征的混合神经网络情感分析模型[J]. 计算机工程与应用, 2025, 61(13): 227-234.

LI Jiaqi, YANG Huan, GAO Hui. Hybrid Neural Network Sentiment Analysis Model Incorporating Character and Word Features[J]. Computer Engineering and Applications, 2025, 61(13): 227-234.

参考文献

[1] HUANG B X, OU Y L, CARLEY K M. Aspect level sentiment classification with attention-over-attention neural networks[C]//Proceedings of the 11th International Conference on Social, Cultural, and Behavioral Modeling. Cham: Springer, 2018: 197-206.
[2] 谭荧, 张进, 夏立新. 社交媒体情境下的情感分析研究综述[J]. 数据分析与知识发现, 2020, 4(1): 1-11.
TAN Y, ZHANG J, XIA L X. A survey of sentiment analysis on social media[J]. Data Analysis and Knowledge Discovery, 2020, 4(1): 1-11.
[3] LING M J, CHEN Q H, SUN Q, et al. Hybrid neural network for Sina Weibo sentiment analysis[J]. IEEE Transactions on Computational Social Systems, 2020, 7(4): 983-990.
[4] DEVLIN J, CHANG M W, LEE K, et al. BERT: pre-training of deep bidirectional transformers for language understanding[J]. arXiv:1810.04805, 2019.
[5] KIM Y, LI P, HUANG H. Convolutional neural networks for sentence classification[J]. arXiv:1408.5882, 2014.
[6] 任菊香, 刘忠宝. 融合多粒度语义特征的中文情感分析方法[J]. 华东师范大学学报(自然科学版), 2023(6): 95-107.
REN J X, LIU Z B. Integrating multi-granularity semantic features into the Chinese sentiment analysis method[J]. Journal of East China Normal University (Natural Science), 2023(6): 95-107.
[7] 衡红军, 杨鼎诚. 融合词法句法信息的方面级情感分析模型[J]. 计算机工程与设计, 2024, 45(3): 837-844.
HENG H J, YANG D C. Aspect-level sentiment analysis model incorporating lexical and syntactic information[J]. Computer Engineering and Design, 2024, 45(3): 837-844.
[8] ZHANG L, WANG S, LIU B. Deep learning for sentiment analysis: a survey[J]. Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery, 2018, 8(4): e1253.
[9] GAN C Q, WANG L, ZHANG Z F, et al. Sparse attention based separable dilated convolutional neural network for targeted sentiment analysis[J]. Knowledge-Based Systems, 2020, 188: 104827.
[10] MA Y, PENG H, CAMBRIA E. Targeted aspect-based sentiment analysis via embedding commonsense knowledge into an attentive LSTM[C]//Proceedings of the 32nd AAAI Conference on Artificial Intelligence (AAAI-18), the 30th Innovative Applications of Artificial Intelligence (IAAI-18), and the 8th AAAI Symposium on Educational Advances in Artificial Intelligence (EAAI-18), 2018: 5876-5883.
[11] ABDELGWAD M M, SOLIMAN T H A, TALOBA A I, et al. Arabic aspect based sentiment analysis using bidirectional GRU based models[J]. Journal of King Saud University-Computer and Information Sciences, 2022, 34(9): 6652-6662.
[12] 于游, 付钰, 吴晓平. 中文文本分类方法综述[J]. 网络与信息安全学报, 2019, 5(5): 1-8.
YU Y, FU Y, WU X P. Summary of text classification methods[J]. Chinese Journal of Network and Information Security, 2019, 5(5): 1-8.
[13] JING H, YANG C. Chinese text sentiment analysis based on transformer model[C]//Proceedings of the 2022 3rd International Conference on Electronic Communication and Artificial Intelligence. Piscataway: IEEE, 2022: 185-189.
[14] JOZEFOWICZ R, VINYALS O, SCHUSTER M, et al. Exploring the limits of language modeling[J]. arXiv:1602.02410, 2016.
[15] 王素格, 李德玉, 魏英杰. 基于赋权粗糙隶属度的文本情感分类方法[J]. 计算机研究与发展, 2011, 48(5): 855-861.
WANG S G, LI D Y, WEI Y J. A method of text sentiment classification based on weighted rough membership[J]. Journal of Computer Research and Development, 2011, 48(5): 855-861.
[16] HEMMATIAN F, SOHRABI M K. A survey on classification techniques for opinion mining and sentiment analysis[J]. Artificial Intelligence Review, 2019, 52(3): 1495-1545.
[17] ZHANG M, ZHANG Y, CHE W, et al. Chinese parsing exploiting characters[C]//Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2013: 125-134.
[18] ABID F, ALAM M, YASIR M, et al. Sentiment analysis through recurrent variants latterly on convolutional neural network of Twitter[J]. Future Generation Computer Systems, 2019, 95: 292-308.
[19] GAN C Q, FENG Q D, ZHANG Z F. Scalable multi-channel dilated CNN-BiLSTM model with attention mechanism for Chinese textual sentiment analysis[J]. Future Generation Computer Systems, 2021, 118: 297-309.
[20] WANG B K, SHAN D H, FAN A W, et al. A sentiment classification method of web social media based on multidimensional and multilevel modeling[J]. IEEE Transactions on Industrial Informatics, 2022, 18(2): 1240-1249.
[21] WANG L, MENG Z Q. Multichannel two-dimensional convolutional neural network based on interactive features and group strategy for Chinese sentiment analysis[J]. Sensors, 2022, 22(3): 714.