A novel cascade hybrid many-objective recommendation algorithm incorporating multistakeholder concerns

Most previous studies of recommender systems (RSs) have particularly focused on optimizing user experience; however, users are not the only stakeholders of an RS. A pure concentration of users limits the ability to incorporate the perspectives of other stakeholders, such as providers. Furthermore, because users' preferences and providers' objectives may conflict, considering only users' views degrades the recommendation methods' utility. Therefore, we propose a cascade hybrid many-objective recommendation method (CHMAOR) to balance four objectives for two different stakeholders. CHMAOR combines provider coverage (PC), user reach coverage (URC), and provider entropy (PE) to create a new provider visibility model (PCRE). The many-objective optimization (MOP) stage includes a novel multiparent probabilistic heuristic genetic algorithm (MPPHX) that heuristically considers both parents' gene frequency and recommendation list features. Extensive experiments demonstrate the following. 1) CHMAOR effectively balances user and provider objectives in terms of accuracy, diversity, novelty, and provider visibility according to the baseline algorithms. 2) The PCRE model considers not only provider coverage but also provider appearance frequency and provider diversity while effectively changing imbalanced provider recommendations. Furthermore, PCRE dramatically reduces the complexity of high-dimensional many-objective recommendations. 3) Our MPPHX achieves better convergence and diversity solutions than the competing MOP algorithms. (c) 2021 Elsevier Inc. All rights reserved.

Automated summary

Previous studies have shown that focusing excessively on users can limit a recommender system's ability to incorporate other stakeholders, while considering only user preferences can degrade the system's utility. Therefore, a new multi-objective recommendation method has been proposed to balance the objectives of both users and providers. Experimental results demonstrate significant improvements in accuracy, diversity, novelty, and provider visibility with this approach.