سازوکار نرخ یادگیری تطبیقی مبتنی بر احساسات برای سیستم‌های توصیه‌گر عصبی در محیط‌های یادگیری تقویت‌شده با فناوری

نوع مقاله : مقاله پژوهشی

نویسندگان

1 دانشجوی دکتری گروه مهندسی کامپیوتر، واحد یزد، دانشگاه آزاد اسلامی، یزد، ایران

2 استاد گروه مهندسی کامپیوتر، دانشگاه یزد، یزد، ایران

3 دانشیارگروه مهندسی کامپیوتر، واحد یزد، دانشگاه آزاد اسلامی، یزد، ایران

10.22034/abmir.2026.24443.1232

چکیده

در اکوسیستم‌های یادگیری تقویت‌شده با فناوری، تغییرات پویای علایق یادگیرندگان که به‌عنوان رانش مفهوم شناخته می‌شود و ناتوانی مدل‌های توصیه‌گر سنتی در انطباق سریع با این تغییرات، چالشی بنیادین محسوب می‌شود. اگرچه سیستم‌های توصیه‌گر عصبی در مدل‌سازی تعاملات غیرخطی موفق بوده‌اند؛ اما استفاده از نرخ‌های یادگیری ایستا یا سازوکارهای کاهش‌دهنده استاندارد در این مدل‌ها، منجر به‌کندی در واکنش به تغییرات ناگهانی سلیقه کاربر می‌گردد. این پژوهش با هدف غلبه بر این محدودیت، یک سازوکار نرخ یادگیری تطبیقی نوین ارائه می‌دهد که در آن نرخ به‌روزرسانی وزن‌های مدل، تابعی مستقیم از شدت و قطبیت احساسات کاربر است که توسط مدل زبانی BERT استخراج می‌شود. فرضیه اصلی این است که احساسات منفی سیگنالی قوی برای نیاز به "اصلاح سریع" مدل هستند. نتایج ارزیابی تجربی بر روی مجموعه‌دادگان مرتبط نشان می‌دهد که این رویکرد، نرخ انطباق مدل با تغییرات ناگهانی سلیقه را تا ۱۳ درصد نسبت به مدل پایه پالایش همکارانه عصبی افزایش داده و ضمن ارتقای دقت رتبه‌بندی، سرعت همگرایی سیستم را در محیط‌های پویا به‌طور معناداری بهبود می‌بخشد.

کلیدواژه‌ها

موضوعات


عنوان مقاله [English]

A Sentiment-Driven Adaptive Learning Rate Mechanism for Neural Recommender Systems in Technology-Enhanced Learning

نویسندگان [English]

  • Mohammad Mehran Lesan Sedgh 1
  • Alimohammad Latif 2
  • Sima Emadi 3
1 PhD student, Department of Computer Engineering, Ya. C., Islamic Azad University, Yazd, Iran
2 Professor, Department of Computer Engineering, Yazd University, Yazd, Iran
3 Associate Professor, Department of Computer Engineering, Ya. C., Islamic Azad University, Yazd, Iran
چکیده [English]

In technology enhanced learning ecosystems, the dynamic evolution of learners’ interests—commonly referred to as concept drift—and the inability of traditional recommender models to rapidly adapt to these changes constitute a fundamental challenge. Although neural recommender systems have demonstrated strong capability in modeling complex nonlinear interactions, the use of static learning rates or conventional decay mechanisms often leads to slow responsiveness to abrupt shifts in user preferences.To address this limitation, the present study proposes a novel adaptive learning rate mechanism in which the weight update rate of the model is directly governed by the intensity and polarity of user sentiments extracted using a BERT based language model. The central hypothesis is that negative sentiments act as a strong signal indicating the need for rapid model correction.Experimental evaluation on relevant datasets shows that the proposed approach improves the model’s adaptation rate to sudden preference changes by up to 13% compared with a baseline Neural Collaborative Filtering model. In addition, the method enhances ranking accuracy while significantly improving system convergence speed in dynamic environments.

کلیدواژه‌ها [English]

  • Educational Recommender Systems
  • Adaptive Learning Rate
  • Neural Collaborative Filtering
  • Sentiment Analysis
  • Deep Learning
  • Concept Drift
[1]     L. Shi, "The integration of advanced AI-enabled emotion detection and adaptive learning systems for improved emotional regulation," J. Educ. Comput. Res., vol. 63, no.1, pp. 173–201, 2025.
[2]     R. Karthika, V. E. Jesi, M. S. Christo, L. J. Deborah, A. Sivaraman, and S. Kumar, "Intelligent personalised learning system based on emotions in e-learning," Pers. Ubiquitous Comput., vol. 27, no.6, pp. 2211–2223, 2023.
[3]     Y. Wang, "Research on online learner modeling and course recommendation based on emotional factors," Sci. Program., vol. 2022, no. 1, p.,2022.
[4]     H. Ezaldeen, R. Misra, S. K. Bisoy, R. Alatrash, and R. Priyadarshini, "A hybrid E-learning recommendation integrating adaptive profiling and sentiment analysis," J. Web Semant., vol. 72, p.,2022.
[5]     S. Abakarim, S. Qassimi, and S. Rakrak, "Emotion and sentiment enriched decision transformer for personalized recommendations," Sci. Rep., vol. 15, no.1,2025.
[6]     S. Li et al.,"Region-aware neural graph collaborative filtering for personalized recommendation," Int. J. Digit. Earth, vol. 15, no.1, pp. 1446–1462, 2022.
[7]     X. Fan, Y. Ji, and B. Hui, "A dynamic preference recommendation model based on spatiotemporal knowledge graphs," Complex Intell. Syst., vol. 11, no. 1, p.46,2025.
[8]     J. Wu, Y. Xu, B. Zhang, Z. Xu, and B. Li, "Graph-based Dynamic Preference Modeling for Personalized Recommendation," in Pacific-Asia Conference on Knowledge Discovery and Data Mining, 2024, pp. 356–368.
[9]     Y. Kim, Y. Lee, V. Yuan, A. Lee, and Y. Lee, "A temporal graph network framework for dynamic recommendation,"arXivPrepr.arXiv2403.16066,2024.
[10] L. Xia, C. Huang, Y. Xu, and J. Pei, "Multi-behavior sequential recommendation with temporal graph transformer," IEEE Trans. Knowl. Data Eng., vol. 35, no.6, pp. 6099–6112, 2022.
[11] T. Sargazi Moghadam, A. Darejeh, M. Delaramifar, and S. Mashayekh, "Toward an artificial intelligence-based decision framework for developing adaptive e-learning systems to impact learners’ emotions," Interact. Learn. Environ., vol. 32, no.7, pp. 3665–3685, 2024.
[12] T. Hussain, L. Yu, M. Asim, A. Ahmed, and M. A. Wani, "Enhancing e-learning adaptability with automated learning style identification and sentiment analysis: a hybrid deep learning approach for smart education," Information, vol. 15, no. 5, p.277,2024.
[13] Y. Tian, S. Peng, X. Zhang, T. Rodemann, K. C. Tan, and Y. Jin, "A Recommender System for Metaheuristic Algorithms for Continuous Optimization Based on Deep Recurrent Neural Networks," IEEE Trans. Artif. Intell., vol. 1, no.1, pp. 5–18, 2021.
[14] E. Dang, Z. Hu, and T. Li, "Enhancing collaborative filtering recommender with prompt-based sentiment analysis," arXiv Prepr. arXiv2207.12883,2022.
[15] K. Zhang et al., "Sifn: A sentiment-aware interactive fusion network for review-based item recommendation," in Proceedings of the 30th ACM International Conference on Information & Knowledge Management, 2021, pp. 3627–3631.
[16] Karabila, N. Darraz, A. El-Ansari, N. Alami, M. Lazaar, and M. El Mallahi, "Recommendation system using Deep Learning-based sentiment analysis," in 2023Sixth International Conference on Vocational Education and Electrical Engineering, 2023, pp. 41–47.
[17] H. Li, J. Zheng, B. Jin, and H. Zhu, "Adaptive Knowledge Contrastive Learning with Dynamic Attention for Recommender Systems," Electronics, vol. 13, no. 18, p. 3594, 2024.
[18] J. C. Salazar, J. Aguilar, J. Monsalve-Pulido, and E. Montoya, "A generic architecture of an affective recommender system for e-learning environments," Univers. Access Inf. Soc., vol. 23, no. 3, pp. 1115–1134, 2024.
[19] Bhatia et al., "A Sentiment Analysis-Based Recommender Framework for Massive Open Online Courses Toward Education 4.0," in Lecture Notes in Networks and Systems, 2023, vol. 421, pp. 817–827.
[20] Y. Cui, H. Yu, X. Guo, H. Cao, and L. Wang, "RAKCR: Reviews sentiment-aware based knowledge graph convolutional networks for Personalized Recommendation," Expert Syst. Appl., vol. 248, p. 123403, 2024.