摘要
为了解决推荐系统的冷启动和数据稀疏性问题,研究人员利用用户之间的信任关系提出了多种基于信任的协同推荐算法,这些方法提高了推荐覆盖率,然而推荐精确度却有所降低。综合考虑用户之间的信任关系和用户的潜在特征,提出了基于信任和概率矩阵分解的协同推荐算法,通过融入用户的相似性、影响力、专业性等知识计算用户之间不对称的信任关系,结合概率矩阵分解模型进行评分预测。最后在数据集上进行实验测试评估,实验表明该算法可以有效提高推荐结果的精确度。
To overcome the problem of cold-start and data sparsity in collaborative fihering recommender systems, the resear- chers utilized the trust relationship between users t6 propose a variety of trust-based recommender algorithms. Though they im- proved the recommender coverage, the recommender precision came down. So this paper took the users＇ influence and the latent factors into account, proposed a trust-based and probabilistic matrix factorization for collaborative filtering recommendation algo- rithm. First, the algorithm integrated the knowledge of the users＇ trust, similitude specialty, and so on, calculated asymmetrical trust value between users. Then it fused the probabilistic matrix factorization method to predict the ratings. Finally, it experi- mented on the real dataset. And the result shows that this algorithm can effectly improve accuracy of rating prediction.
出处
《计算机应用研究》
CSCD
北大核心
2016年第11期3240-3244,共5页
Application Research of Computers
基金
国家自然科学基金资助项目(61171108)
国家"973"计划资助项目(2012CB315901
2012CB315905)
国家科技支撑计划资助项目(2014BAH30B01)
关键词
推荐系统
协同过滤
信任
数据稀疏
冷启动
矩阵分解
recommender systems
collaborative filtering
trust
data sparsity
cold-start
matrix factorization
