Purpose Learning stakeholders’ preferences for personalized recommendations is challenging in the multi-stakeholder recommendation system. Existing fairness-based recommendation approaches address this issue but overlook stakeholders’ implicit preferences regarding item characteristics and their latent relationships, thereby failing to produce a balanced trade-off recommendation environment. Hence, this study aims to learn users’ implicit preferences on various items’ aspects to generate fair and trade-off recommendations. Design/methodology/approach This work develops a deep autoencoder and a similarity reinforcement-based multi-criteria recommendation system (DeepSRMCRS) to evaluate criterion-wise similarity and forecast the overall rating using aggregated similarity. The proposed multi-stakeholder utility optimization-based recommendation framework creates stakeholders’ personalized utility functions and optimizes conflicting utilities using evolutionary optimization. An extensive simulation and comparative evaluation using a multi-criteria rating-based real-world healthcare data set validates the proposed models’ performance over baseline methods. Findings The proposed model improves the utility score by up to 24% for patient, doctor and pharma stakeholders and 26% for the system stakeholders across real-time and benchmark datasets. The DeepSRMCRS exhibits superior predictive performance improvement of up to 11% in MAE, 13% in RMSE and 9% in F1-score. The proposed model also attains an 18% and 21% improvement in exposure and hit score. Originality/value A distinctive, fair and equitable information retrieval approach to healthcare recommendation systems combining deep learning with multi-criteria similarity reinforcement and multi-stakeholder utility optimization.
Shrivastava et al. (Mon,) studied this question.