基于协同过滤技术的电商用户重复购买行为预测模型构建

Construction of E-commerce Users'Repeated Purchase Behavior Prediction Model Based on Collaborative Filtering Technology

由于电商平台用户浏览方向随机性较强,感兴趣信息量巨大,对用户重复购买行为的预测难度较高。为增强用户体验感与满意度,提高电商服务质量,构建基于协同过滤技术的电商用户重复购买行为预测模型。获取电商用户历史数据集,从用户-商品、用户-商品品类两个层面提取与用户重购行为具有强关联性的统计、衍生、行为衰减特征,将其输入到XGBoost算法中,预测用户是否有重复购买行为。采用协同过滤算法建立重复购买行为用户的评分矩阵,通过Pearson相关系数衡量两用户间的相似度,确定重购行为用户的最近邻集合。根据最近邻用户对目标商品的评分计算重购行为用户对目标商品的评分,获取评分高的数个商品生成推荐列表,实现电商用户重购行为预测。实验结果表明:该模型能够实现重购行为用户的预测,CART树数量为500、特征量为60、推荐列表长度为3时,该预测模型预测性能最突出,推荐结果的用户评分可达9.8。

Due to the strong randomness of users'browsing directions on e-commerce platforms and the huge amount of information they are interested in,it is difficult to predict users'repeated purchases.In order to enhance user experience and satisfaction,and improve the quality of e-commerce service,a prediction model of e-commerce users'repeated purchase behavior based on collaborative filtering technology was built.Obtain the historical data set of e-commerce users,extract the statistical,derivative and behavioral attenuation features that are strongly related to users'repurchase behavior from the user commodity and user commodity categories levels,and input them into the XGBoost algorithm to predict whether users have repeated purchase behavior.The collaborative filtering algorithm is used to establish a scoring matrix for users with repeat purchase behavior.The Pearson correlation coefficient is used to measure the similarity between the two users and determine the nearest neighbor set of users with repeat purchase behavior.According to the score of the nearest neighbor user on the target goods,the user's score on the target goods is calculated,and several goods with high scores are obtained to generate a recommendation list,so as to realize the prediction of e-commerce users'repurchase behavior.The experimental results show that the model can achieve the prediction of users with repurchase behavior.When the number of CART trees is 500,the feature quantity is 60,and the length of the recommendation list is 3,the prediction performance of the prediction model is the most outstanding,and the user rating of the recommendation results can reach 9.8.