基于改进用户画像的协同过滤推荐算法

Collaborative Filtering Recommendation Algorithm Based on Improved User Profile

由于科技的发展,协同过滤算法也在不断优化。上述算法在个性化推荐系统的设计中较为常用,其当前存在最大的问题就是数据稀疏。结合基本特征得到的用户画像,能够有效避免数据稀疏的问题。针对传统协同过滤算法的稀疏性问题,提出了一种基于用户画像的协同过滤推荐算法(CPCF),可以保证其在用户规模扩大的同时保持推荐的高效性和准确性。CPCF算法创新在于将用户评分矩阵转化为用户特征矩阵,提出一种改进的用户画像,并采用用户画像与传统相似度方法相结合生成CPsim相似度,最后使用改进的权重聚合方法进行评分估计,通过逆序排序实现推荐过程。在两个数据集上的实验结果显示,相比UBCF等算法(MAE)降低了13%,精确率、召回率和F1-Score分别提高了11%、3%和4%,可见结合协同过滤算法与用户画像进行推荐系统的设计,可有效提升系统精准度。

In recent years, collaborative filtering algorithm has been widely used in personalized recommendation systems and achieved good results, but the problem of data sparsity is still not well solved. However, a user portrait can avoid the sparsity in a user feature matrix by extracting the user’s basic features. To relieve this sparsity problem, we propose a collaborative filtering recommendation algorithm(CPCF) based on user portraits, which can ensure high efficiency and accuracy of recommendations while the user scale expands. The innovation of the CPCF algorithm is to transform the user rating matrix into the user feature matrix, propose an improved user portrait, and combine the user portrait with the traditional similarity method to generate CPsim similarity. Finally, the improved weight aggregation method is used for rating estimation, and the recommendation process is realized through reverse-order sorting. The experimental results on two datasets show that compared with UBCF and other algorithms(MAE),the accuracy, recall and F1 score are improved by 13%,11%,3% and 4%,respectively. It can be seen that the design of the recommendation system combined with the collaborative filtering algorithm and user portrait can effectively improve the accuracy of the system.