Khan Hikmat Ullah, Naz Anam, Alarfaj Fawaz Khaled, Almusallam Naif
Department of Information Technology, University of Sargodha, Punjab, Pakistan.
Department of Management Information Systems, School of Business, King Faisal University, Al Ahsa, Saudi Arabia.
Sci Rep. 2025 Jul 8;15(1):24503. doi: 10.1038/s41598-025-08931-1.
Recommender systems are considered powerful tools, in the modern era, for filtering a huge amount of information and delivering personalized content, particularly in domains like e-commerce, social media, and entertainment. In the context of movie recommendations, accurately modeling user preferences based on past interactions, ratings, and contextual metadata is crucial for enhancing user satisfaction. With the rising trends and influence of Artificial Intelligence (AI), advanced models are increasingly being employed to enhance the precision and adaptability of such systems. This study proposes a novel transformer-based architecture, MetaBERTTransformer4Rec(MBT4R), designed to outperform state of the art existing methods in the relevant literature. The extensive empirical analysis is carried out on two datasets which based on the same source, publicly available known as MovieLens, which is a standard dataset for movie recommendation. The proposed model utilizes a self-attention mechanism to effectively capture sequential dependencies and contextual relationships, enabling deeper understanding of users' preferences. The results reveal that MBT4R achieves the lowest RMSE of 0.62, MAE of 0.45, and highest R² of 0.39, significantly superior to the benchmarks established by traditional models including machine learning (DT, KNN, RF, XGB), matrix factorization (SVD) and deep learning (GRU). This research highlights the effectiveness of AI techniques in improving the accuracy and personalization of recommendation systems focusing on enhancing user satisfaction by accurately predicting user preferences and delivering tailored film suggestions. It also provides a pathway for future advancements in personalized user experiences across entertainment platforms.
在现代，推荐系统被视为强大的工具，用于筛选海量信息并提供个性化内容，尤其是在电子商务、社交媒体和娱乐等领域。在电影推荐的背景下，基于过去的交互、评分和上下文元数据准确建模用户偏好对于提高用户满意度至关重要。随着人工智能（AI）的兴起及其影响力的不断扩大，先进的模型越来越多地被用于提高此类系统的精度和适应性。本研究提出了一种基于Transformer的新型架构，即MetaBERTTransformer4Rec（MBT4R），旨在超越相关文献中现有的最先进方法。在两个基于相同来源的公开可用数据集（即MovieLens，这是电影推荐的标准数据集）上进行了广泛的实证分析。所提出的模型利用自注意力机制有效地捕捉序列依赖性和上下文关系，从而能够更深入地理解用户偏好。结果表明，MBT4R实现了最低均方根误差（RMSE）为0.62、平均绝对误差（MAE）为0.45以及最高决定系数（R²）为0.39，显著优于传统模型（包括机器学习（决策树、K近邻、随机森林、极端梯度提升）、矩阵分解（奇异值分解）和深度学习（门控循环单元））所建立的基准。这项研究突出了人工智能技术在提高推荐系统准确性和个性化方面的有效性，通过准确预测用户偏好并提供量身定制的电影建议来提高用户满意度。它还为跨娱乐平台的个性化用户体验的未来发展提供了一条途径。
