MULTILINGUAL SENTIMENT UNDERSTANDING OF SOCIAL MEDIA TEXT STREAMS WITH A TRANSFORMER
Abstract
The mining of social media information through sentiment Sentiment analysis has emerged as a significant research issue because of the fast increase in multilingual user-generated content on social media, ecommerce portals, and online forums. Even though transformer-based language models have greatly enhanced the performance of text classification, most of the current systems use a single model across all languages, which might not be the best with linguistically diverse inputs. Besides that, efficiency and interpretability of deployment are frequently ignored in the real-world sentiment analysis pipelines. The current paper introduces CAHTS, a Context-Adaptive Hybrid Transformer Framework, which is a multilingual sentiment analysis framework with an efficient real-time inference. The suggested approach is a hybrid of RoBERTa to English-centric sentiment prediction and XLM-RoBERTa to multilingual text comprehension via a Dynamic Language Routing Mechanism (DLRM). The routing module uses language characteristics to choose the most appropriate encoder pathway based on the characteristics detected. A Context-Adaptive Attention Fusion (CAAF) module is used to combine sentence-level and token-level embeddings and then classify. The framework has four sentiment categories, which include Positive, Negative, Neutral, and Mixed. ONNX-based quantization is used to optimize model deployment to achieve lower latency and memory
footprint without compromising competitive accuracy. Ablation analysis also reveals that routing and attention fusion can be measured to bring about improvements. The proposed system provides a viable tradeoff between accuracy, multilingual scalability and deployment efficiency, and is applicable in applications like customer feedback analytics, social media monitoring, and multilingual conversational systems.
Downloads
References
COVID-19 Tweets Using Deep Learning Techniques. IEEE Access, 8,
209707-209716.
[2] Zhang, Y., Feng, X., & Yan, X. (2020). Sentiment Analysis of Chinese
social media Using Deep Reinforcement Learning. IEEE Access, 8,
136821-136831.
[3] Arifin, A. Z., Dhandapani, P. W., & Peregrinate, A. (2021). Sentiment
Analysis of Indonesian social media Using a Hybrid Approach.
International Journal of Advanced Computer Science and Applications,
12(4), 190-198.
[4] Zhang, Y., Feng, X., & Yan, X. (2021). Dynamic Emotion Detection
and Multi-task Learning for Sentiment Analysis of Microblogs. IEEE
Transactions on Computational Social Systems, 8(1), 190-199.
[5] Bojarski, P., & Sikora, M. (2021). Sentiment Analysis of Polish
Tweets Using Deep Learning and Rule-based Approaches. Journal of
Universal Computer Science, 27(2), 126-146.
[6] Jbene, M., Tigani, S., Rachid, S., & Chehri, A. (2018). Deep Neural
Network and Boosting Based Hybrid Quality Ranking for e Commerce
Product Search. IEEE Access, 6, 21930-21940. doi:
10.1109/ACCESS.2018.2822925.
[7] Dey, T., Poria, S., & Cambria, E. (2021). Multi-task
learning for
sentiment analysis on social media. Information Processing &
Management, 58(1), 102387.
[8] Lee, J. Y., Kim, J. W., Lee, J. H., & Lee, J. G. (2021).
Online
sentiment analysis with trend information on social media.
Future
Generation Computer Systems, 120, 463-472.
[9] Kumar, S., Sharma, R., & Sharma, N. (2021). Sentiment
Analysis of COVID-19 Tweets: A Comparative Study of
Deep Learning and Traditional Machine Learning
Approaches. Cognitive Computation, 13(5), 5769-5782.
[10] Chen, Y., Liu, Y., Liu, S., & Zhang, Q. (2021). A novel
approach for cross-domain sentiment analysis on social media
using recurrent neural networks. Expert Systems with
Applications, 168, 114339.
[11] Guo, W., Zhang, J., & Zhao, Y. (2021). Sentiment
Analysis of Chinese social media Based on Ensemble
Learning. IEEE Access, 9, 103610-103619.
[12] Wang, Y., Xu, Y., Liu, Y., Yang, X., & Wu, C. (2021).
A hybrid approach for sentiment analysis on social media using
machine learning and knowledge graph. Knowledge- Based
Systems, 232, 107125.
[13] Jiang, H., & Qiu, X. (2021). A Bi-directional Long Shortterm Memory with Attention Mechanism for Sentiment
Analysis in Chinese Microblog. Journal of Ambient
Intelligence and Humanized Computing, 1-10.
[14] Liu, Y., Zhang, C., Gao, H., & Huang, W. (2021).
Sentiment analysis on social media using a hybrid deep
learning approach. Journal of Ambient Intelligence and
Humanized Computing, 12(8), 8257-8271.
[15] Lu, Y., & Wang, W. Y. (2021). Sentiment analysis of
COVID-19 related tweets using deep learning and transfer
learning. Journal of Biomedical Informatics, 118, 103792.
[16] Solakidis, G. S., Vavliakis, K. N., & Mitkas, P. A. (2014).
Multilingual sentiment analysis using emoticons and keywords.
In IEEE/WIC/ACM International Conference on Web
Intelligence and Intelligent Agent Technology (WI-IAT).
[17] Denecke, K. (2008). Using SentiWordNet for multilingual
sentiment analysis. In IEEE International Conference on Data
Engineering Workshops (ICDE Workshops).
[18] Baliyan, A., Batra, A., & Singh, S. P. (2021). Multilingual
sentiment analysis using RNN-LSTM and neural machine
translation. In 8th International Conference on Computing for
Sustainable Global Development (INDIACom).
[19] F. Barbieri et al., “XLM-T: Multilingual Language Models
in Twitter for Sentiment Analysis,” in LREC, 2022.
[20] J. Howard and S. Ruder, “Universal Language Model
Fine-tuning for Text Classification,” in ACL, 2018.
[21] M. Peters et al., “Deep Contextualized Word
Representations,” in NAACL, 2018.
Copyright (c) 2026 IJRDO -Journal of Computer Science Engineering
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Author(s) and co-author(s) jointly and severally represent and warrant that the Article is original with the author(s) and does not infringe any copyright or violate any other right of any third parties, and that the Article has not been published elsewhere. Author(s) agree to the terms that the IJRDO Journal will have the full right to remove the published article on any misconduct found in the published article.
