融入用户长短期兴趣的推荐算法多样性优化
2021-10-24夏瑞玲李国平王国中
夏瑞玲 李国平 王国中



摘要: 多数传统的推荐算法在追求准确度时,忽略了多样性也是衡量推荐效果不可或缺的指标之一。而一味地提升多样性又势必会造成准确度的极大损失。由此提出依据用户兴趣度和兴趣变化度,在计算出用户兴趣值的基础上,分析不同用户的兴趣偏好情况。再将用户的长期与短期兴趣相结合进行推荐,保障个性化的同时确定用户的多样化程度,生成最终的推荐列表,很好地平衡了推荐结果的准确度与和多样性。
关键词: 长短期兴趣; 准确度; 多样性; 推荐系统
中图分类号:TP391 文献标识码:A 文章编号:1006-8228(2021)10-44-05
Diversity optimization of recommendation algorithm integrating
user's long-short-term interest
Xia Ruiling1, Li Guoping2, Wang Guozhong2
(1. College of Electrical and Electronic Engineering, Shanghai University of Engineering Science, Shanghai 201620, China;
2. Key Laboratory of Artificial Intelligence Application State Administration of Radio and Television)
Abstract: In the pursuit of accuracy, most traditional recommendation algorithms ignore diversity, which is also one of the indispensable indicators to measure the recommendation effect. However, increasing diversity desperately is bound to result in a great loss of accuracy. Presented on considering user interest and interest changes, on the basis of calculating user interest value, analyze the different user interest preference. Then combining the user's long-term and short-term interest to recommend, guaranteeing the diversification of personalized determine a user at the same time, produce the final recommendation list, reached a good balance between accuracy and diversity of recommendations.
Key words: long-short-term interest; precision; diversity; recommendation system
0 引言
在網络信息蜂拥而至的今天,推荐系统的诞生,无疑为信息的生产者和消费者提供了一套解决信息过载问题的双赢方案[1]。推荐列表需要覆盖的不单是某一种兴趣领域,对不同的场景,所应用的推荐方式不尽相同,但就其推荐目标来说都具有根本的一致性:提升用户体验,吸引并留住用户。而多样性是衡量推荐列表中物品属性、种类的差异性程度[2]。也就是推荐列表中的物品单调与否。多样而丰富的推荐结果会让用户更满意,推荐内容多样性越好,用户的长期留存的概率越大。
基于用户的协同过滤算法通过同类用户分析,对目标用户进行推荐,一定程度上提升了多样性但推荐结果趋于热门化[3]。从长远看,要发掘用户的长期需求,也要精准满足短期需求,覆盖用户绝大多数的兴趣。因此,人们一直都在利用推荐系统的种种优势探索和拓展用户兴趣,在保证准确率的前提下,尽量提升推荐内容的多样性。……
