Many famous Internet companies possess tens of millions of daily active users, who would generate massive row logs with features such as big data, diversity and disorder, including browse, click, favorite and buy logs when they are online. Big data increases the complexity and time when information consumers try to get valid informa-tion from information provider which will deteriorate the user experience. To address the problem, in this study, a plan-ning method is proposed when considering recommend accuracy and time. In order to solve the planning problem, an recommend system using cluster algorithm before factorization machine recommending is proposed. Experimental re-sults show that the proposed method can perform recommendation a more accurate and low-delay way.%随着电商行业的飞速发展，电商平台上产生的点击、评分、购买等行为日志数据朝着海量化、多样化、无序化的方向发展，使得获取有价值信息的复杂度增加，降低了信息生产者将信息传达给信息消费者的效率，使得用户体验变差。为了解决上述问题，本文提出一种基于用户行为聚类的分级因子分解机系统推荐模型。首先构建用户多维行为特征工程，将用户分为四种行为模式，随后对不同模式的用户分别采用因子分解机推荐算法进行推荐预测。最后仿真结果表明，本文提出的改进推荐算法不仅降低了推荐预测的均方根误差（RMSE），并且大大地减少了推荐时间，有利于实时推荐系统的应用。
展开▼
