基于信任关系和项目流行度的矩阵分解推荐算法

针对现有推荐系统推荐覆盖范围不高的问题,提出一种融合项目流行度和用户信任关系的矩阵分解推荐算法。合并用户-项目评分矩阵和用户-用户信任关系矩阵,通过矩阵分解的方式同时传递信任和推荐项目,极大提高了推荐算法的覆盖率,但损失了现有方法8%左右的精度。将项目流行度作为权重因子,引入到高稀疏性的用户-项目评分矩阵中,根据项目流行度对用户评分项目和未评分项目分别进行加权处理,提高了推荐算法的准确率。通过在Epinions数据集上的对比实验结果表明,该算法在大幅度改善推荐覆盖率的同时,保证了推荐的准确率,能够给于用户更好的推荐效果。

MATRIX FACTORIZATION RECOMMENDATION ALGORITHM BASED ON TRUST RELATIONSHIP AND ITEM POPULARITY

To solve the problem of low coverage of existing recommendation systems, we proposed a matrix factorization recommendation algorithm that combined item popularity and user trust relationship. We merged the user-item scoring matrix and the user-user trust relationship matrix to transfer trust and recommendation items simultaneously through matrix factorization. It greatly improved the coverage of the recommended algorithm, but lost accuracy up to 8%. We introduced the item popularity as a weighting factor into the high sparse user-item rating matrix. According to the item popularity, we weighted user rating items and non-rating items respectively, which improved the accuracy of the recommendation algorithm. The experiment results on the Epinions dataset show that our algorithm can greatly improve the recommended coverage rate without reducing the accuracy of the recommendation, so it can give users better recommendation results.