【摘要】 目前协同过滤的主流方法是矩阵分解模型。针对传统矩阵分解方法没有考虑用户偏见和物品隐含特征对推荐质量的共同影响,在矩阵分解模型的基础上提出了一种基于用户偏见修正的联合矩阵分解算法(联合分解物品评分矩阵和物品共现矩阵)。在不同基准数据集上的实验结果反映了所提策略的合理性,并通过基于排序的指标证明了所提模型比传统矩阵分解模型在性能上有较大幅度的提升。更多还原

【Abstract】 Although matrix factorization model has become the major method in the collaborative filtering,it ignores the combined influence of the user bias and latentitems characteristics on recommendation quality.Therefore,this research proposed a collective matrix factorization algorithm,which factorizes items rating matrix and items co-occurrence matrix to amend user bias based on matrix factorization model.The experimental results from different benchmark datasets prove the rationality of the combined factorization algorithm,and indicate greater improvement in the ranking-based metrics in comparison with the traditional matrix factorization model.更多还原