基于特征增益与多级优化的协同过滤个性化推荐算法  

A Recommendation Algorithm Based on Collaborative Filtering by Feature Augmentation and Cascade Tactics

作　　者：马胡双 石永革[1] 高胜保[2] 

机构地区：[1]南昌大学信息工程学院计算机科学与技术系,南昌330038 [2]中国电信股份有限公司江西分公司,南昌330029

出　　处：《科学技术与工程》2016年第21期272-277,共6页Science Technology and Engineering

基　　金：国家自然科学基金(61163005);江西省科技计划项目(2014ZBBE50008)资助

摘　　要：基于混合策略个性化推荐的思想,为进一步提升业务套餐型产品的个性化推荐的准确性,提出了基于特征增益与多级优化的协同过滤推荐算法(FACCF);其中融合了消费数据的时域特征、空域特征、消费倾向以及套餐特征。该算法首先基于客户的时域、空域行为特征,构建了CTAP概率主题模型实现协同过滤;其次,将过滤后的新特征、套餐主题与套餐特征结合进行优化;最后,基于贝叶斯网络对客户群体的消费倾向进行聚类分析,形成二次优化,获得个性化推荐列表。使用电信真实用户数据实证分析的结果表明,FACCF算法能够更准确地预测客户消费意愿。Most personalized recommendation based on user and products, without combined with environmental factors and physical features, so that the algorithm can＇ t be a good predictor of consumer preferences. This paper presents a mixed strategy of personalized recommendation algorithm FACCF a CTAP （ Customer-Time-Area Pack-age） probabilistic topic model, where the customer behavior fusion domain features, spatial characteristics, pay tendencies and product features; The algorithm is based on the timedomain, area-domain collaborative filtering characteristics of the spatial behavior; after filtration new features, product themes and product features to optimize the combination; later, based on Bayesian network customer groups pay a tendency to cluster analysis, the formation of secondary optimization, get a personalized recommendation list. Use real customer behavior data to conduct empirical analysis results show that the FACCF algorithm show more accurately predict the customer, s consumption, and enhance personalized recommendation accuracy.

关 键 词：业务套餐推荐 协同过滤 特征增益 多级优化 概率主题模型 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]