一种嵌入项目元数据的跨项目协同过滤推荐算法

Cross-item Collaborative Filtering Recommendation Algorithm for Embedded Item Metadata

传统的协同过滤推荐系统中用户相似性通常是由余弦、Pearson等方法来计算.然而这些方法都是基于共同的评分项目,在评分数据稀疏的情况下,用户之间的共同评分项目较少,很难准确地找到与目标用户偏好相似的邻居.针对这一问题,提出一种新的方法,通过嵌入项目元数据,利用用户与项目及其属性的交互,采用贝叶斯概率计算项目特征的相似性,并结合EM D方法实现基于元数据表示的跨项目计算相似性.实验结果表明,该方法能够有效地解决用户评分数据稀疏情况下传统的相似性度量方法所存在的不足,提高了推荐的准确性.

In traditional collaborative filtering recommender systems,User similarity is usually calculated by cosine similarity or Pearson correlation coefficient.However,these methods are based on common ratings,in the case of rating datas sparse,there is less common scoring between users,and it is difficult to accurately find neighbors that are similar to the target user＇s preferences.In order to solve this problem,we propose a new method by embedding the item metadatas,using the interaction between the user and the item and its attributes.Bayesian probability is used to calculate the similarity of the project characteristics,and the EMD method is used to realize the similarity of the cross item computation based on metadata representation.The experimental results show that the proposed method can effectively solve the shortcomings of the traditional similarity measure method in the case of sparse user data,thus improving the accuracy of the recommendation.