An Adaptive Probabilistic Evolutionary Multi-task Algorithm for Multi-objective Recommendation

Recommender systems have become immensely popular and widely used currently. Multi-objective evolutionary algorithms (MOEAs) have demonstrated successful applications in this field by taking into account both accuracy and various other performance metrics at the same time. To address the issue of high computational complexity brought about by large-scale recommendations, the concept of evolutionary multi-task (EMT) has been successfully applied. This study further presents a novel EMT-based algorithm named APEMA, which incorporates a new adaptive probabilistic operator that utilizes a variety of historical information from the population for comprehensive evaluation. By guiding the genetic operator and knowledge transfer, it can accelerate the convergence of results while simultaneously increasing the diversity of the population, thereby achieving better performance. Specifically, the APEMA algorithm uses its unique probabilistic operator to adapt to the evolving population. The effectiveness of APEMA in addressing multi-objective recommendation problems is empirically validated on the Movielens and Douban datasets. Experimental results show that it achieves superior performance compared to several state-of-the-art algorithms.