Deep learning-based modeling of stakeholders preferences and balancing through multi-objective optimization in a multi-stakeholder recommendation system

Satisfying the stakeholders' preferences and establishing a trade-off recommendation environment is a challenging problem in a multi-stakeholder recommendation system. Recent techniques have maximized consumer satisfaction by delivering diverse and novel item suggestions but do not satisfy provider and system-level goals. The multi-stakeholder recommendation methods exploit highly sparse user-item interaction to compute providers' and systems' utility and deliver recommendations without establishing trade-offs among stakeholders' conflicting preferences. It hinders their comprehension of stakeholders' preferences and minimizes the likelihood of item consumption, providers' exposure, and stakeholders' retention in the system. Therefore, this study proposes a Deep learning-based Multi-Stakeholder recommendation model and develops a personalized Consumption, Exposure, and Retention maximization-based objective function for consumer, provider, and system stakeholders. This work employs a deep autoencoder to learn latent implicit features of stakeholder-item interactions. Next, we propose a deep neural network-based probabilistic modeling of the objective functions that maximize the likelihood of item consumption, providers' exposure, and stakeholders' retention. Finally, this study employs evolutionary multi-objective optimization to establish the trade-off recommendation solution between stakeholders' conflicting objectives. The simulations and evaluation results using standard evaluation metrics for each stakeholder across four benchmark datasets validate the proposed models' performance over baseline methods.