一种巴氏系数改进相似度的协同过滤算法

摘要

Aiming at the problem of low-quality recommendation and data sparsity, we proposed a collaborative filtering algorithm based on improved similarity measure with Bhattacharyya coefficient.First, we use Jaccard similarity to calculate the global similarity between users based on neighbor cooperative filtering algorithm.Secondly, we use the Bhattacharyya coefficient to obtain the whole law of the grade distribution.And we combine the Pearson correlation coefficient to calculate the local similarity.Finally, we fuse the global similarity and local similarity to obtain final similarity metric.The experimental results show that algorithm can get better recommendation results on sparse data sets.It effectively mitigates the sparseness of scoring data and improves the recommended accuracy.%针对传统协同过滤算法中评分数据稀疏性及所造成推荐质量不高的问题,提出一种巴氏系数(Bhattacharyya Coefficient)改进相似度的协同过滤算法.在基于近邻协同过滤算法基础上,首先利用Jaccard相似性来计算用户间的全局相似性;其次使用巴氏系数获得评分分布的整体规律,并结合Pearson相关系数来计算其局部相似性;最后融合全局相似性和局部相似性得到最终的相似度矩阵.实验结果表明,该算法在稀疏数据集上获得更好的推荐结果,有效地缓解了评分数据稀疏性问题,提高了推荐的准确度.