结合多信息的概率矩阵分解模型

Probability Matrix Factorization Model Combined with Multiple Information

为了改善传统协同过滤推荐算法的冷启动与数据稀疏问题,基于概率矩阵分解模型,将用户属性、物品关系与时序行为融合到模型中,通过不断调整3种模型所占权重,得到最小的RMSE值。在Movielens数据集上进行实验,并与其它相关算法的RMSE值进行比较。实验结果表明,结合多信息的概率矩阵分解模型的RMSE值低于其它推荐方法,即推荐精度优于其它方法。结合多信息的概率矩阵分解模型,在数据稀疏情况下,也能保持较好的推荐性能,推荐精度得到一定程度提升。

In order to improve the cold start and data sparsity of the traditional collaborative filtering recommendation algorithm,user attributes,item relationship and sequential behavior are merged into probabilistic matrix factorization model,and the minimum RMSE value is obtained by constantly adjusting the weight of the three models.Experiments on the Movielens data set are compared with the RMSE values of other related algorithms.The experimental results show that the RMSE value of the probability matrix factorization model with multiple information is lower than other recommended methods,and recommendation accuracy is better than the other methods.The probability matrix factorization model combined with multi information can also maintain better recommendation performance in the case of data sparsity,and the accuracy of recommendation is improved.