摘要
在信息过载的时代,推荐系统通过分析用户的历史行为,为用户兴趣建模,主动给用户推荐能够满足他们兴趣和需求的信息,已经被广泛应用于电子商务等多个领域。但是在推荐系统中,用户评分数据极端稀疏,矩阵的稀疏性导致推荐算法在相似性计算时存在较大误差,进而导致最近邻居选择的不准确,从而影响推荐质量。针对上面存在的问题,文中通过对评分矩阵采用PCA降维的方法,降低了评分矩阵的稀疏性,保留了最能代表用户兴趣的维数,使得相似性计算更加准确,保证了最近邻居选择的准确性,从而提高了推荐质量。实验结果表明,在公开数据集上与传统的协同过滤推荐算法相比较,文中提出的算法具有较高的准确度和覆盖度。
In the era of information overload,recommender system can help users find their interest and recommend the satisfactory information to analyze their historical behavior,so it is widely used in electronic commerce and other fields. But the user rating matrix is extremely sparse in recommender systems. The sparsity of the matrix leads to great error in the calculation of similarity of recommendation algorithms,bringing about the nearest neighbor sections is not accurate,thus affecting the quality of recommendation. Aiming at the problems above,a dimension reduction method based on PCA was proposed to reduce the sparsity of user rating matrix,by this method the remain matrix retain the most representative characteristic of the user interest,so that the similarity calculation is more accurate to ensure the accuracy of the nearest neighbors,thereby improving the quality of the recommendation. The experimental results show that compared with the traditional collaborative filtering algorithm,the algorithm proposed reaches a high accuracy and coverage.
出处
《计算机技术与发展》
2016年第2期26-30,共5页
Computer Technology and Development
基金
国家自然科学基金资助项目(41271387)
陕西师范大学院士创新基金资助项目(999521)
西安市科技计划基金资助项目(SF1228-3)
关键词
主成分分析
降维
协同过滤
推荐算法
PCA
dimension reduction
collaborative filtering
recommendation algorithm