结合隐语义模型与改进用户信任的协同过滤推荐算法

Collaborative filtering recommendation algorithm combining latent factor model and improved customer trust

为了改善传统的协同过滤推荐算法由于数据稀疏性以及过于强调相似度的作用而导致预测准确率较低的现象,提出一种改进算法,即结合隐语义模型与改进用户信任的协同过滤推荐算法。该算法首先利用隐语义模型对原始评分矩阵进行空缺值填充,从而降低矩阵的稀疏度;其次在传统的用户信任关系上引入正负反馈因子来降低用户信任误差,并将改进的用户信任通过线性加权的方式与修正余弦相似度相结合,从而改善用户之间的相似性,使得计算结果更具实际意义。在MovieLens数据集上进行仿真,结果表明,所提算法能有效地提高推荐的预测准确性。

As the traditional collaborative filtering recommendation algorithm has low prediction accuracy due to data sparseness and overemphasis on similarity,an improved algorithm,that is,collaborative filtering recommendation algorithm combining latent factor model and improved customer trust,is proposed.In the algorithm,the latent factor model is used to fill the original scoring matrix with vacancy values to reduce the sparseness of the matrix,and then the positive and negative feedback factors are introduced in the traditional user trust relationship to reduce user trust error.The improved user trust is combined with the correction cosine similarity in linear-weight mode to improve the similarity among users.This makes the results more practical.The result of simulation on the MovieLens data set shows that the proposed algorithm can effectively improve the prediction accuracy of recommendation.