Abstract: With the rapid development of cloud computing and mobile Internet technique,the amount of cloud offers and online information has been growing explosively,which yields information overload.To deal with the data sparsity problem and the cold start problem in recommender systems,an intelligent collaborative recommendation method based on spectral clustering and latent factor model was proposed.Similar users were clustered with the spectral clustering scheme according to the label features of users.The original rating matrix was transformed into multiple lowdimensional sub-matrix where the factorization-based latent factor model was employed to predict the missing data locally.Afterwards,the final predictions could be made globally based on the improved neighbor-based collaborative recommendation algorithm.The proposed method was effective in dealing with the data sparsity problem and the cold start problem.Experimental results validated that the proposed method was improved in recommendation accuracy and efficiency.

Key words: collaborative recommendation, spectral clustering, latent factor model, matrix factorization

摘要： 随着云计算及移动互联网技术的迅速发展,网络中可选服务信息呈爆炸性增长,信息过载问题日益严重。针对推荐系统中存在的数据稀疏性问题及冷启动问题,提出一种基于谱聚类和隐语义模型的智能协同推荐方法。该方法基于提取的用户标签特征信息,利用谱聚类算法对相似用户进行聚类,将原始高维评分矩阵转化为多个较低维的子评分矩阵。然后在子评分矩阵中利用隐语义模型对缺失评分进行局部预测。最后在获得缺失评分后利用改进的基于邻域的协同推荐算法对目标用户进行全局评分预测。所提算法有效解决了数据稀疏性问题和冷启动问题,在提高预测准确度的同时加快了推荐算法效率。

关键词: 协同推荐, 谱聚类, 隐语义模型, 矩阵分解