基于用户相似度的加权项目偏差SlopeOne协同过滤推荐算法

摘要

A collaborative filtering algorithm was designed to help make a reasonable prediction of targeted users’personalized preferences in the current study.It could be achieved through digging out the behavior of the user population with common experience from massive historical data.Businesses could then supply customized products and information recommendation for the targeted users according to these predictions, which were found significant and valuable.Traditional SlopeOne collaborative filtering algorithm is simple and high-efficiency,but its accuracy is relatively low.Therefore,a novel weighted item deviation algorithm was proposed in order to further improve the predicting accuracy,and retain the efficiency advantage of tra-ditional algorithm.The weighted item deviation SlopeOne collaborative filtering recommendation algorithm was obtained based on the user similarities (WID_SlopeOne_US)after optimizing measure of the deviation between items.Lots of experimental results had shown that our new algorithm present higher accuracy,ef-ficiency and stability.%协同过滤算法旨在从海量的历史数据中，挖掘出拥有共同经验的用户群体的行为习惯，以此来协助对目标用户的个性化偏好作出合理的预测。根据这些预测结果，对目标用户进行有针对性的产品或资讯推荐，这对商家来说具有重大的意义和价值。传统的 SlopeOne协同过滤推荐算法虽然实现简单，运行效率高，但其准确率不高。为了进一步提高预测结果的准确率，同时又尽可能的保留原算法所具有的效率优势，提出了基于用户相似度的加权项目偏差计算方法，优化了项目之间偏差性的衡量尺度，从而得到基于用户相似度的加权项目偏差 SlopeOne协同过滤推荐算法（WID_SlopeOne_US）。大量实验证明，新算法具有更高的准确率、较高的效率、和良好的稳定性等优点。