用户与项目语义相似性填充的协同过滤推荐方法

Collaborative Filtering Recommendation Method Based on User and Item Semantic Similarity Filling

为解决协同过滤推荐方法中矩阵的稀疏性导致推荐精度下降的问题,文中提出一种结合用户属性及项目特征的矩阵填充协同过滤推荐方法。通过用户属性特征计算用户间语义相似性,构建项目领域本体以计算项目间语义相似性,加权用户语义相似性和项目语义相似性预测出的评分值填充评分矩阵,并获取近邻用户进行推荐。在MovieLens数据集上利用平均绝对误差和接收者工作特性(ROC)曲线两个评价标准进行实验验证,结果表明:相较于传统的协同过滤方法(TCF)、基于缺省值填充的协同过滤方法(CFBDF)和基于均值填充的协同过滤方法(CFBAF)三种方法,本文方法的平均绝对误差(MAE)值更低,ROC曲线下方的面积(AUC)值更大,这说明文中方法可以有效缓解矩阵的稀疏性,进而提高推荐的准确性。

The sparsity of collaborative filtering method will result in poor recommendation quality.In order to improve the recommendation accuracy,a collaborative filtering recommendation method is proposed which combines user attributes and item characteristics.Firstly,the user semantic similarity is calculated through user attribute characteristics,and the item semantic similarity is calculated by ontology.Then,the rating matrix is filled with weighted scores which are predicted by user semantic similarity and item semantic similarity.Finally,recommendation is conducted after the neighbors are obtained.The average absolute error and ROC curve calculated by the MovieLens dataset show that,compared with the TCF,CFBDF and CFBAF,the MAE value of this methed is lower and the ACU value is larger,which shows that this method can effectively reduce sparsity and improve the recommendation accuracy.