摘要
为了解决传统商品推荐方法仅考虑商品两两相似性或只通过商品属性的简单集成构建推荐网络图,对网络对象复杂性和依赖关系考虑不够导致推荐准确性低的问题,提出一种改进的商品推荐算法.算法通过商品、品牌、店铺及关联关系构建混合图,根据节点关系、节点出度、商户广告付费和商品点击数构建数学模型,得到商品、品牌和店铺间的转移概率,建立节点初始概率转移矩阵.通过重启动随机游走算法确定最终节点概率转移矩阵,实现商品推荐.实验结果表明,与当下常用推荐算法相比,该算法提高了商品推荐的准确率(Precision);算法扩展性强,适用于各种电商平台.
To solve the problem of low accuracy of traditional recommendation algorithm caused by only consider the commodity two similarity or only through simple integration build commodity attribute recommendation network diagram,the network object complex- ity and dependency consider not comprehensive, an improved e-commerce recommendation algorithm is proposed. Algorithm by commodity,brand,shop and the relationship between them building heterogeneous network diagram,according to the relationship between node, the node degrees, merchants paid advertising and clicks to build mathematical model, get the transfer probability between the commodity, brand and shop ,initial probability transfer matrix is established. Through the restart random walk algorithm to determine the probability of the final node transfer matrix, to recommend commodities. The experimental results show that compared with the commonly used algorithm,the proposed algorithm improves the accuracy (Precision) of the recommendation;algorithm scalability for in a variety of electronic business platform.
出处
《小型微型计算机系统》
CSCD
北大核心
2016年第11期2433-2436,共4页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61562041)资助
江西省自然科学基金项目(20142bab217009)资助
关键词
混合图
随机游走
商品推荐
转移矩阵
F值
hybrid graph
random walk
advertising recommendation
transfer matrix
F-measure