基于客户Web时空行为轨迹的兴趣点预测方法

Method of interest points prediction based on customer web temporal behavior trajectory

客户兴趣点预测是大数据环境下提高电子商务推荐精度的关键,针对现有客户兴趣预测未综合考虑客户多种行为和时序时间的影响问题。为研究一种基于客户Web时空行为轨迹的兴趣点预测方法,构建了包含客户、时间、行为和兴趣点四层子网的客户Web时空行为超网络模型,并引入行为影响因子,提出基于超边相似性的兴趣点预测算法,在建立连通矩阵的基础上,通过邻接矩阵计算、超三角形判定和超边相似度计算,得到相似度最高的超边,该超边对应的兴趣点即为预测结果。实验结果表明,该方法在时间误差允许范围内,兴趣点预测准确度随时间精度的减小而增加,与传统的标签预测方法相比,预测准确度由56.2%提高至74%。

Interest point is the key to improving the accuracy of e-commerce recommendation under big data environment. However, the existing predictive research ignores the comprehensive impact of various customers＇ behaviors and time series on interest points. In order to make up this gap, the article sets up a customer Web space and time super network model which involves four subnets： customer, time,behavior and interest point, and establishes the influence factors of behavior. Then, based on similarity of superlink prediction method and the establishment of connectivity matrix, the adjacency matrix is calculated and super triangle judgement is made, so that the most similar super edge and the best prediction results of interest points are obtained. Finally, experiment shows that the precision of interest prediction gets better with the decrease of time accuracy within the allowable range of time error. Compared with the traditional method of label prediction, the prediction accuracy is improved from 56.2% to 74%.