基于TimeSVD++与DPC的推荐算法研究

Research on Recommendation Algorithm Based on TimeSVD++and DPC

针对使用奇异值分解(SVD)方法时需要填充矩阵的内容较多以及K-means算法受到K值的影响且数据集形状限制等问题,提出一种将TimeSVD++与改进的密度峰值聚类结合的方法。首先在SVD++的基础上引入参数时间因子,构建TimeSVD++模型;其次,采用将相似系数引入高斯核函数的方法,对密度峰聚类算法中的局部密度公式进行修正;引入信息熵确定最优截断距离,最后在数据集MovieLens-1M和MovieLens-100k上验证,并将实验结果与其它算法进行对比。结果表明:所提出的方法在MAE,RMSE,Recall和F1值指标上均优于其它的算法。

In view of the problems that the singular value decomposition method needs to fill the matrix too much,the k-means algorithm is affected by the K value and the shape of the data set is limited,this paper proposes a method combining TimeSVD++with the improved density peak clustering.Firstly,time factor was introduced to construct TIMESVD++model based on SVD++.Secondly,by introducing the similarity coefficient into the Gaussian kernel function,the local density formula in the density peak clustering algorithm was modified.Information entropy was introduced to determine the optimal truncation distance,and finally the data sets MovieLens-1M and MovieLens-100k were verified.The experiments showed that the proposed method was superior to other algorithms in MAE,RMSE,Recall and F1 value indexes.