基于机器学习的融合推荐算法研究

Research on Hybrid Recommendation Algorithm Based on Machine Learning

为了缓解信息爆炸的困境,采用机器学习算法建立一个融合的推荐系统以提高预测准确性和聚合推荐多样性。针对稀疏的数据集及推荐结果单一的问题,提出了以协同过滤为基础的天牛须搜索优化的交替最小二乘法模型、基于密度的噪声应用空间聚类的用户聚类模型、并建立了XGBoost融合排序模型,从而得到个性化推荐。采用来自亚马逊平台的苹果手机销售数据,对三个模型进行仿真测试,结果表明:与单一的交替最小二乘法相比新模型拓展性高,收敛速度快,具有更好的实用价值。

In order to alleviate the dilemma of information explosion, a fused recommendation system is built to improve the accuracy of prediction and aggregate the diversity of recommendations by machine learning algorithm. According to the problems of sparse data sets and single recommendation results, the alternative-least-square optimization model of Beetle antennae search algorithm based on collaborative filtering and the user clustering model based on density-based spatial clustering of noise application are presented. And the XGBoost fusion sorting model is built to get personalized recommendation. The three models are simulated with the sales data of Apple iPhone from Amazon platform. The results show that compared with the single alternating least squares method, the new model has high expansibility, fast convergence and better practical value.