大数据时代基于用户需求的推荐多样性方法研究综述

Literature Review of Recommendation Diversity Methods Based on Users’Demands in the Era of Big Data

现在推荐多样性方法能够在一定程度上满足用户的需求,但大数据时代,仍面临多样性感知差异问题、相关性与多样性平衡问题和多样性性能提升问题。通过从提高推荐多样性所运用的关键理论和技术出发,剖析推荐结果多样化的过程,本文探讨了大数据时代下推荐系统中经典的推荐多样性方法及逻辑思路,并总结出主要提高推荐多样性的方法,且对其性能特点和局限性进行详细分析,最后基于推荐多样性方法面临的挑战进行总结和展望。本文经过梳理后,将推荐多样性方法主要总结为以下四个:基于长尾效应的推荐多样性方法、基于二部图的推荐多样性方法、基于行列式点过程的推荐多样性方法和基于注意力机制的推荐多样性方法。

Nowadays,recommendation diversity methods can meet users’demands to a certain extent,but in the era of big data,there are still problems of diversity perception differences,relevance and diversity balance and diversity performance improvement.By analyzing the process of diversifying recommendation results from the key theories and technologies used to improve recommendation diversity,this paper discusses the classical recommendation diversity methods and logical paths in recommendation systems in the era of big data,and summarizes the main methods to improve recommendation diversity,and analyzes the characteristics and limitations of performance in detail,and finally summarizes and predicts the challenges faced by recommendation diversity methods.After sorting out,this paper summarizes the recommendation diversity methods into the following four,including recommendation diversity methods based on the long tail effect,based on the bipartite graph,based on the deterministic point process(DPP),and based on attention mechanisms.