基于用户概要扩展的协同过滤算法

Collaborative filtering algorithm based on expanding user profile

针对协同过滤算法中的新用户冷启动问题,提出了基于用户概要扩展的协同过滤算法(EUPCF)。算法采用一种新的加权朴素贝叶斯方法对新用户的概要进行局部扩展,然后使用扩展后的概要为新用户进行预测推荐,为预测项目提供更多近邻项目。新的加权朴素贝叶斯方法为每个条件属性独立计算后验概率,避免了传统方法中联合分布先验概率对数据稀疏度的敏感性问题,提高了扩展的准确度。Movie Lens数据集实验表明,新算法拥有良好的预测准确度,同时,不会对推荐的实时性产生较大影响。

In order to alleviate cold-start problem the new user faced by collaborative filtering algorithm,this paper proposed a new collaborative filtering algorithm based on expanding user profile( EUPCF). The new algorithm firstly used a new weight naive Bayesian method to get local expanding of the new user's profile,and then used the expanded profile to predict and recommend for the new user,which could provide more accurate neighbor items for the to-be-predicted items. The new weighted naive Bayesian method in EUPCF did not use joint distribution and calculated posteriori probability for each condition attribute independently,which avoided the sensitive problem of sparse data faced by joint distribution. The comparative experiments on Movie Lens dataset show that the proposed EUPCF algorithm has better performance on the prediction accuracy and has slight hindering effect on real-time recommendation.