基于相似度优化偏差计算的slope-one算法研究

Study on Slope-one Algorithm Based on Similitude Optimization Deviation Calculation

slope-one算法是个性化推荐系统中最简洁的协同过滤推荐算法,常用于评分预测来进行矩阵填充从而降低原始数据的稀疏性。由于传统slope-one算法在计算偏差时考虑了所有评分项目,而将不相关的项目纳入偏差计算反而会降低预测的准确性,文章针对该问题提出一种改进的slope-one算法,先通过项目相似度筛选出待预测评分项目的近邻用户集S,在用户集S上计算项目间的偏差值dev时引入用户相似度,从而有效地提高了评分预测的可靠性。在Movielens-1M数据集上对本文算法和slope-one算法(SO)以及以用户相似度为权重的slope-one算法(BUW-SO)作五折交叉实验,结果表明,改进的算法不仅能减少时间和空间复杂度,还能提高预测的准确性,使推荐系统有更好的推荐效果。

The slope-one algorithm is the most concise collaborative filtering recommendation algorithm in personalized recommendation system.It is often used for scoring prediction to fill the matrix so as to reduce the sparsity of the original data.As the traditional slope-one algorithm takes all the scoring items into account in calculating the deviations,and the inclusion of unrelated items in the deviation calculation will actually reduce the accuracy of forecasting,this paper proposes an improved slope-one algorithm.Firstly,the near neighbor user set S of the items to be predicted and graded is screened out through item similitude,and then user similitude is introduced when calculating the deviation value dev between items on user set S.Thus,the reliability of scoring prediction is effectively improved.The five-fold crossover experiment is performed on the movielens-1 M data set for the algorithm in this paper,the slope-one algorithm(SO)and the slope-one algorithm(BUW-SO)weighted by user similitude.The results show that the improved algorithm can not only reduce time and space complexity,but also improve the accuracy of prediction,making the recommendation system have better recommendation effect.