| 摘要: |
| 为有效解决传统推荐算法精度低的问题，提出了一种融合用户偏好和社交活跃度的概率矩阵分解推荐算法(Probabilistic Matrix Factorization Recommendation Algorithm Combining User Preference and Social Activity，UPSA-PMF)，通过用户评分数据计算用户间的偏好信任度时，使用了共同项目平衡因子和热门项目惩罚因子进行改进；计算社交网络中的信任度时，考虑了社交活跃度与用户信任度的关系，并将社交活跃度作为惩罚因子，修正用户信任度。将偏好信任度和社交网络中的信任度以动态组合的方式得到最终的信任度，将最终的信任度与概率矩阵模型相结合，实现推荐。实验证明，改进的算法均优于现有的推荐算法，有效提高了推荐质量。 |
| 关键词: 社交网络 概率矩阵分解 用户偏好 社交活跃度 评分预测 |
| DOI： |
|
| 基金项目:国家自然科学基金资助项目（61671095） |
|
| A recommendation algorithm combining user preference and social activity |
| LI Lingling,HUANG Jun,WANG Yue |
| (School of Communication and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China) |
| Abstract: |
| To effectively solve the problem of poor accuracy of traditional recommendation algorithm,a probabilistic matrix factorization recommendation algorithm combining user preference and social activity(UPSA-PMF) is proposed.When calculating the preference trust between users through user rating data,the common item balance factor and the popular item penalty factor are used for improvement.When calculating the trust in social networks,the relationship between social activity and user trust is considered,and social activity is used as a penalty factor to modify user trust.The preference trust degree and the trust degree in the social network are dynamically combined to obtain the final trust degree.The final degree of trust is combined with the probability matrix model to realize the recommendation.Experiments show that the improved algorithms are better than the existing recommendation algorithms,which effectively improves the quality of recommendations. |
| Key words: social network probabilistic matrix factorization user preference social activity score prediction |
