摘要
研究用户优化服务算法问题,应为用户提供个性化的推荐服务的系统。Top-N推荐问题,是指通过对用户历史偏好信息的挖掘,给每个用户推荐N个最可能喜好的内容。针对上述问题,提出了一种面向排序的推荐算法EIBRO-MF,通过融合系统中的显式和隐式反馈数据,建立用户喜好的偏序对关系来训练协同过滤的参数模型,最后利用优化的模型参数给出推荐结果。仿真结果表明,与传统的协同过滤算法、以及只能利用隐式反馈数据的排序算法相比,提出的算法能大幅提高推荐列表的排名精准度。
Recommender system uses customers＇ historical preference data to provide personalized recommendation. Top-N recommendation problem is that systems recommend to each user N items that they probably like most. To address this problem, we proposed a ranking oriented algorithm EIBRO-MF. Firstly we blended explicit and implicit feedback data to construct user-item preference pairs. Then we used the preference pairs to train a collaborate filtering model and obtained the recommendation results at last. Experiments with real-world data sets demonstate that the proposed algorithm can greatly improve the ranking precision of recommendation lists in constract with alternative methods like traditional CF-based algorithm and ranking algorithm.
出处
《计算机仿真》
CSCD
北大核心
2013年第5期264-268,共5页
Computer Simulation
关键词
推荐系统
面向排序
显示反馈
隐式反馈
Recommender system
Ranking oriented
Explicit feedback
Implicit feedback
