基于GEP-RBF的协同过滤数据稀疏性问题研究

Data Sparse Issue of Collaborative Filtering Algorithm Based on GEP-RBF

针对传统协同过滤推荐算法的数据稀疏性问题,提出了基于GEP-RBF的协同过滤推荐算法。该算法对目标用户偏好的分类范畴进行了分析,构建了局部用户-项目评分矩阵,同时利用GEP优化RBF神经网络,预测局部用户-项目评分矩阵的缺失评分,平滑评分矩阵,并给出了用户评分项目交集阈值修正相似度的方法,提高用户相似度计算的准确性。实验结果表明,该算法能有效地缓解数据稀疏性问题,从而提高了协同过滤推荐系统的推荐质量。

Aiming at the problems of the data sparse in traditional collaborative filtering algorithm, the collaborative filtering algorithm based on GEP-RBF is presented. The category of the target user preference is analyzed, the local user-item scoring matrix is built, and Radi- cal Basis Function neural network is optimized by gene expression programming. The value of null rating in the matrix is predicted by opti- mized radical basis function neural network for smoothing rating matrix. A method of user rating items intersection threshold to correct simi- larity is provided for improving the accuracy of user similarity computing. The results showed that the algorithm could alleviate the problem of the data sparse, so as to improve recommendation quality of collaborative filtering recommender systems.