An Intelligent Framework for Persian Sentiment Analysis in Digital Marketing Using Transformer-Based Models
محورهای موضوعی : Artificial Intelligence Tools in Software and Data Engineeringparinaz shahrokhi fard 1 , vahid Torkzadeh 2
1 - Department of Computer and Information Technology, Islamic Azad University, Mashhad, Iran.
2 - Department of Computer and Information Technology, Faculty of Engineering, Islamic Azad University, Mashhad, Iran.
کلید واژه: Sentiment Analysis, Digital Marketing, Deep Learning, Transformer Model, Persian Natural Language Processing.,
چکیده مقاله :
Extended Abstract
Introduction & Problem Statement: The rapid growth of e-commerce has turned user-generated text into a vital resource for understanding customer attitudes. However, Persian sentiment analysis faces significant hurdles due to linguistic ambiguities, complex morphology, and a lack of large-scale labeled datasets. There is a notable research gap regarding a localized, integrated framework that bridges technical modeling with practical digital marketing applications.
Research Objective & Hypotheses: This study proposes an intelligent framework based on the ParsBERT architecture to enhance the accuracy and stability of sentiment detection in marketing environments. The core hypothesis is that a hybrid approach—combining deep contextual representations from Transformers with traditional statistical features—will outperform baseline methods like LSTM and Naive Bayes.
Methodology: The research follows a five-stage process: targeted data collection of 20,000 samples from Persian Twitter and Digikala, specialized Persian text preprocessing (normalization and noise removal) , and the development of a hybrid model. This model concatenates the ParsBERT [CLS] token embedding with TF-IDF statistical feature vectors at the representation level before passing them to a Softmax classifier.
Findings: Comparative evaluations demonstrate that the proposed ParsBERT-based framework achieves superior results with a Precision of 0.90, Recall of 0.88, and F1-score of 0.89. The model achieved an overall accuracy of 90%. Data analysis revealed a sentiment distribution of 54% positive, 31% negative, and 15% neutral among Persian users in the marketing dataset.
Conclusion: The study concludes that integrating native language models like ParsBERT with hybrid feature sets significantly improves decision-making in digital marketing. The framework effectively reduces classification errors caused by Persian's complex structure and provides actionable insights for brand reputation and customer experience management.
Keywords: Sentiment Analysis, Digital Marketing, Deep Learning, Transformer Model, Persian Natural Language Processing.
Extended Abstract
Introduction & Problem Statement: The rapid growth of e-commerce has turned user-generated text into a vital resource for understanding customer attitudes. However, Persian sentiment analysis faces significant hurdles due to linguistic ambiguities, complex morphology, and a lack of large-scale labeled datasets. There is a notable research gap regarding a localized, integrated framework that bridges technical modeling with practical digital marketing applications.
Research Objective & Hypotheses: This study proposes an intelligent framework based on the ParsBERT architecture to enhance the accuracy and stability of sentiment detection in marketing environments. The core hypothesis is that a hybrid approach—combining deep contextual representations from Transformers with traditional statistical features—will outperform baseline methods like LSTM and Naive Bayes.
Methodology: The research follows a five-stage process: targeted data collection of 20,000 samples from Persian Twitter and Digikala, specialized Persian text preprocessing (normalization and noise removal) , and the development of a hybrid model. This model concatenates the ParsBERT [CLS] token embedding with TF-IDF statistical feature vectors at the representation level before passing them to a Softmax classifier.
Findings: Comparative evaluations demonstrate that the proposed ParsBERT-based framework achieves superior results with a Precision of 0.90, Recall of 0.88, and F1-score of 0.89. The model achieved an overall accuracy of 90%. Data analysis revealed a sentiment distribution of 54% positive, 31% negative, and 15% neutral among Persian users in the marketing dataset.
Conclusion: The study concludes that integrating native language models like ParsBERT with hybrid feature sets significantly improves decision-making in digital marketing. The framework effectively reduces classification errors caused by Persian's complex structure and provides actionable insights for brand reputation and customer experience management.
Keywords: Sentiment Analysis, Digital Marketing, Deep Learning, Transformer Model, Persian Natural Language Processing.
[1] I. K. S. Al-Tameemi et al., “A comprehensive review of visual–textual sentiment analysis from social media networks,” J. Comput. Soc. Sci., vol. 7, pp. 2767–2838, Sep. 2024. [Online]. Available: https://doi.org/10.1007/s42001-024-00326-y.
[2] M. Rodríguez-Ibáñez et al., “A review on sentiment analysis from social media platforms,” Expert Syst. Appl., vol. 223, Art no. 119862, 2023. [Online]. Available: https://doi.org/10.1016/j.eswa.2023.119862.
[3] P. Vijayaragavan et al., “Sustainable sentiment analysis on E-commerce platforms using a weighted parallel hybrid deep learning approach,” Sci. Rep., vol. 14, Art. no. 78318, Nov. 2024. [Online]. Available: https://doi.org/10.1038/s41598-024-78318-1.
[4] F. X. Arias et al., “Sentiment analysis of public social media as a tool for health-related topics,” IEEE Access,vol.10,pp.7583475848,Jul.2022.[Online].Available:https://doi.org/10.1109/ACCESS.2022.3187406.
[5] R. Paul and M. Rodriguez, “AI-powered sentiment analysis in digital marketing: A systematic review,” SSRN Preprint,2023.[Online].Available: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4567890.
[6] N. A. Sharma et al., “A review of sentiment analysis: tasks, applications, and deep learning techniques,” Int. J. Data Sci Anal., vol. 19, pp. 351–388, Jul. 2024. [Online]. Available: https://doi.org/10.1007/s41060-024-00594-x.
[7] A. Daza et al., “Sentiment analysis on E-commerce product reviews using ML and DL algorithms: A bibliometric analysis and systematic review,” Int. J. Inf. Manag. Data Insights, vol. 4, no. 2, Art. no. 100267, Nov. 2024. [Online]. Available: https://doi.org/10.1016/j.jjimei.2024.100267.
[8] X. Chen et al., “A computational analysis of aspect-based sentiment analysis research through bibliometric mapping and topic modeling,” J. Big Data, vol. 12, Art. no. 40, Feb. 2025. [Online]. Available: https://doi.org/10.1186/s40537-025-01068-y.
[9] Y. Mao et al., “Sentiment analysis methods, applications, and challenges: A systematic literature review,” J. King Saud Univ.Comput. Inf. Sci., vol. 36, no. 4, Art. no. 102048, Apr.2024.[Online]. Available: https://doi.org/10.1016/j.jksuci.2024.102048.
[10] M. Selimovic, “The role of sentiment analysis in brand management and marketing,” WJARR, vol. 24, no. 2, 2024. [Online]. Available: https://wjarr.com/sites/default/files/WJARR-2024-2369.pdf.
[11] Izadi, M.M., “Advanced sentiment analysis of Persian E-commerce reviews using ontology and ParsBERT,” in 11th Int, Conf. Electrical, Computer, Mechanical Engineering and Artificial Intelligence, Tehran, Iran, 2025. [Online]. Available: https://civilica.com/doc/2294628/
[12] Jamali, Sh., Azmayandeh, M., and Parastar, B., “Sentiment analysis of review texts using the BERT model,” in 3rd Specialty Seminar on Data Science and Its Applications,Tehran,Iran,2024.[Online].Available: https://civilica.com/doc/2152541/
[13] Norouzi, R.M., “A Transformer-based model for sentiment analysis,” in 7th Int. Conf. Electrical Engineering, Computer Science and Information Technology, Tehran, Iran,2024.[Online].Available: https://civilica.com/doc/1638016/
[14] Ardabili, S., and Akbari, S., “Sentiment analysis through artificial intelligence,” in 23rd Int. Conf. Information Technology, Computer and Telecommunications, Tehran, Iran, 2024. [Online]. Available: https://civilica.com/doc/2059708/
[15] Rajabi, Z., Valavi, M., and Hourali, M., “A review of sentiment analysis methods in Persian texts,” Signal and Data Processing Journal, vol. 19, no. 2, pp. 107–132, Summer2022.[Online].Available: https://www.sid.ir/paper/1099717/fa.
