加入用户评分偏置的推荐系统排名模型

针对现有推荐技术忽视了推荐结果总体多样性的问题,提出一种加入用户评分偏置的推荐系统排名模型.该模型通过交叉验证获取偏置调节参数,用以计算用户评分偏置,利用该偏置改进了传统模型中的用户阈值计算方法,从而保证总体多样性的提升效果可作用于全局用户.通过在阈值计算环节引入多样性调节因子,并结合多种排名算法,实现了总体多样性和精准度的精细控制.实验结果表明,相对于传统的基于经验取值的排名模型,所提模型在保证推荐精准度的同时,提高了推荐结果的总体多样性,可以方便地嵌入互动电视和电子商务等推荐系统,无需对原功能模块进行改动.

A Rank Model with User Bias for Recommender System

A new rank model that takes user’s bias into account is proposed to solve the problem that existing rank techniques do not consider the aggregate diversity of the recommendation result.The bias regulating parameter is obtained by cross validation,and is used to calculate the user’s bias.Then the algorithm to calculate the user’s threshold in the traditional rank models is improved by using the resulting user’s bias,so that the enhancement of the aggregate diversity can be effective to all users.The goal of fine control on the aggregate diversity and accuracy is achieved by taking the diversity regulating parameter into the process of calculating the threshold and combining with multiple rank methods.Experimental results and comparisons with the traditional rank model based on experience show that the proposed model can generate recommendations that have higher aggregate diversity across all users while maintaining the recommendation accuracy.Besides,the proposed model can be embedded into interactive TV or E-commerce recommender system without changing the original functional components.