基于高维稀疏数据聚类的协同过滤推荐算法

Collaborative Filtering Recommendation Algorithm underHigh Sparse Data Clustering

针对协同过滤推荐算法面临数据高维稀疏特征时推荐效果较差的缺点,在现有高维稀疏数据聚类研究的基础上,利用评分数据稀疏差异度和项目类别构造集合差异度度量公式,用以在用户—项目评分矩阵上进行项目聚类。在此基础上进行项目相似性计算和最近邻居查询,然后对用户未评分的项目进行评分预测,进而产生推荐。实验证明本文提出的基于稀疏差异度和项目类别的项目聚类算法及在此基础上的协同过滤推荐结果优于传统的K-means聚类算法基础上的推荐效果。同全项目集协同过滤推荐相比较,在效率和推荐精度上也表现出一定的优越性。

In order to resolve the poor-quality of recommenda tion in collaborative filtering recommendation algorithms in case of the high sparse dataset,this paper proposes a novel algorithm named item-based clustering recormmendation algorithm(IBCRA).One of characteristics in the IBCRA is that it has considered the properties of data sparse difference and item category clustering within user itemn dataset.Specifically,on the basis of high dimensions data clustering algorithms,the IBCRA algorithm uscs the rating data sparsc difference and item categories in the rating dataset to construct a measuring formula for calculating dataset difference,where the formula is used for item clustering in user item rating array.Then the IBCRA calculates item similarity and searches for k nearest neighbors of target item based on the outcome of item clustering.Finally it forecasts the ratings for those no rating item in dataset and so generates recommendations.The experimental results show the IBCRA has improved the recommendation quality in collaborative filtering recommendation.The comparative experiments and parameter sensitivity analysis also show,in perspective of the accuracy and speed of convergence,the IBCRA also outperforms the collaborative filtering recommendation algorithm with all items based algorithm.