Andy Supriyadi, Herman Dwi Surjono, Handaru Jati, Nurul Firdaus
Hybrid deep learning recommender systems have received significant interest from e-commerce recently, mainly because of their potential for improving recommendation quality. However, core issues such as recommendation diversity, scalability, interpretability and real-time adaptability remain largely unexplored in previous works. Also, the trustworthiness of reviews using the ABSA for better product recommendations is not well exploited. This paper systematically reviews hybrid DL models in e-commerce recommendation systems, focusing on the advancement of the models, challenges they face, and ABSA integration. Based on the PRISMA process, SLR is carried out with a total of 691 papers being screened, from which twenty-five were chosen to be reviewed in more detail. The review proves that to combine sentiment analysis with ABSA can highly enhance the performance of score-valued product recommending system both in terms of accuracy and diversity. These hybrid systems resolve some of the most well known problems like cold-start, data sparsity and over-specialization and promotes increased personalization of user experience. Unlike prior work, this overview takes a direct look at the role of ABSA in enhancing product recommendations and provides a different insight on its use. This paper also discusses unaddressed issues pertaining to scalability, interpretability, and online adaptation. In conclusion, this research emphasizes not only the significant advantages but also the existing headwinds of hybrid models in e-commerce and suggests future areas for academic research.. © 2025 IEEE.
Universitas Sebelas, Vocational School, Department of Informatics Engineering, Maret, Surakarta, Indonesia; Universitas Negeri Yogyakarta, Department of Electrical Engineering, Yogyakarta, Indonesia
