摘要
针对传统协同过滤推荐算法推荐精度较低等问题,提出一种基于贝叶斯后验概率预测和非合作博弈的个性化推荐算法。采用文件主题模型求取用户与其参加过的所有社交活动的主题分布,利用隐含主题概率分布表征用户兴趣度,根据信任传递机制求取用户的直接信任和间接信任,形成用户间的信任度;将用户的兴趣度和信任度等隐式特征赋予合理的先验分布,利用贝叶斯后验概率预测隐式特征后的显式反馈;依据显式反馈将推荐结果转化为非合作博弈中用户效益最大化的纳什均衡求解。仿真对比实验表明,与其他三种推荐算法相比该算法的查准率至少提高了3.13%,查全率至少提高了2.62%。
Aimmed at the problems of low accuracy of traditional collaborative filtering recommendation algorithms,a personalized recommendation algorithm based on Bayesian posterior probability prediction and non-cooperative games is proposed.The file topic model was used to get the topic distribution of users and all social activities they have participated in.The implicit topic probability distribution was used to represent the user interest.According to the trust transfer mechanism,the direct trust and indirect trust of users were calculated to form the trust between users.The implicit features such as user interest and trust were given a reasonable prior distribution,and Bayesian posterior probability was used to predict the explicit feedback after the implicit features.Based on the explicit feedback,the recommendation results were transformed into the Nash equilibrium solution of maximizing user benefit in non-cooperative games.The simulation comparison experiments show that compared with the other three recommended algorithms,this algorithm’s accuracy rate is improved by at least 3.13%and its recall rate is increased by at least 2.62%.
作者
索岩
程向羽
Suo Yan;Cheng Xiangyu(Xinlian College,Henan Normal University,Xinxiang 453000,Henan,China;College of Computer and Information Engineering,Henan Normal University,Xinxiang 453000,Henan,China)
出处
《计算机应用与软件》
北大核心
2022年第3期270-276,284,共8页
Computer Applications and Software
基金
河南省软科学研究计划项目(142400411151)。
关键词
贝叶斯后验概率
兴趣度
信任度
非合作博弈
个性化推荐
Bayesian posterior probability
Interest degree
Trust degree
Non-cooperative game
Personalized recommendation
