融合时序的决策树推荐算法研究

Decision Tree Recommendation Algorithms Integrating Time-Sequence

提出融合用户画像的多特征属性来提高性能推荐算法,主要创新在于:①融合用户行为的时序,将用户的历史行为按照时间顺序存储;②收集用户对推荐项目的反馈对目标函数采用机器学习来实时优化预测参数;③基于决策树的方法和优化损失函数更新预测参数。通过在Gowalla、Last.fm两个数据集上与采用多个基本推荐算法的结果相比较,该算法有效地优化推荐系统准确性。

Proposes to improve a recommendation algorithm based on time sequence and decision model of users’actions,the main innovations here are as follows:i.integrating the time sequence of user's behavior,storing the user's historical behavior in time sequence;ii.updating the prediction parameters in the repeated interaction with users and objective function by machine learning;iii.predicting parameters based on decision tree and optimizing loss function.Compared with other intelligent algorithms in the dada set of Gowalla and Last.fm,the algorithm can be more accurate.